Review Luck et al.

– ERP studies of attention

Event-related potential
studies of attention
Steven J. Luck, Geoffrey F. Woodman and Edward K. Vogel

Over the past 30 years, recordings of event-related potentials (ERPs) from normal
individuals have played an increasingly important role in our understanding of the
mechanisms of attention. This article reviews some of the recent ERP studies of attention,
focusing on studies that isolate the operation of attention in specific cognitive subsystems
such as perception, working memory, and response selection. Several conclusions are
drawn. First, under some conditions attention modulates the initial feedforward volley of
neural activity in intermediate visual processing areas. Second, these early effects can be
observed for both the voluntary allocation of attention and for the automatic capture of
attention following a peripheral visual transient. Third, these effects are present not only
when attention is directed to a location in 2-dimensional space, but also when attention is
directed to one of two spatially overlapping surfaces. Fourth, attention does not modulate
sensory activity unless sensory systems are overloaded; when sensory systems are not
taxed, attention may instead operate to influence memory or response processes. That is,
attention operates to mitigate information overload in whichever cognitive subsystems
are overloaded by a particular combination of stimuli and task.

T here has been a great deal of excitement over the past selection’ issue; indeed, this was the first mainstream ques-
decade about the possibility of using the techniques of neuro- tion about attention that ERPs were used to answer1,4.
science to answer fundamental questions about cognition, To assess the locus of selection, one simply compares the
and one of the great success stories has been the use of event- ERP waveform elicited by an attended stimulus to the ERP
related potential (ERP) recordings to study attention. ERPs waveform elicited by a physically identical stimulus when it is
have been used to study attention since the 1960s (Ref. 1), ignored. The earliest time point at which the two waveforms
but conceptual and methodological advancements have led differ provides an upper bound on the initial effect of attention
to a recent surge in ERP studies that provide answers to on the processing of the stimulus (it is an upper bound because
mainstream cognitive questions. The purpose of this article there might be earlier effects that are not evident in the ERP
is to review several of these studies, making them accessible waveforms). For example, Fig. 1 compares two ERP wave-
to a broader audience. Towards this end, the basics of the forms, both elicited by a rectangle presented in the left visual
ERP technique and its advantages for studying cognition are field; one waveform was elicited by this stimulus when atten-
described in Box 1. In this review, we will focus on studies tion directed to the left visual field, and the other was elicited
that have sought to isolate the operation of attention within by this same stimulus when attention was directed to the right
specific cognitive subsystems, but recent ERP studies have visual field. These waveforms begin to differ in the latency
also addressed other important issues, such as the time range of the P1 wave – between 60 and 100 ms poststimu-
course of attentional orienting (as described in Box 2). lus – which indicates that attention modulates the processing
of the stimulus at or before this time. Many studies have shown
Evidence for early selection this general pattern of results5– 9. Because visual information has
S.J. Luck, Perhaps the most fundamental question about attention is just begun to reach the extrastriate visual areas during this time
G.F. Woodman and whether attention modulates information processing at a range10, these results provide strong evidence that attention in-
E.K. Vogel are at the
sensory stage or at a later stage. There are obvious cases in fluences sensory coding, at least under some conditions11.
Department of
which attention operates after a stimulus has been per- These ERP results have been interpreted as a sensory ‘gain
Psychology, University
ceived; for example, we may see a stimulus and simply control’ mechanism that simply causes larger P1 responses for
of Iowa, 11 Seashore
Hall E, Iowa City, IA
choose not to make an overt response to it. It is much more attended-location stimuli relative to ignored-location stimuli12.
52242-1407, USA. difficult to determine whether attention can sometimes However, this interpretation leads to a quandary: if attention
suppress the sensory processing of a stimulus. In fact, it is simply increases the gain of the sensory input, it seems that
tel: +1 319 335 2422 not clear that traditional behavioral methods have ever attention would increase the noise as well as the signal, leading
fax: +1 319 335 0191
e-mail: steven-luck@ yielded unambiguous evidence for early selection2,3. to no improvement in the signal-to-noise ratio. As a solution
uiowa.edu However, ERPs are well suited for addressing this ‘locus-of- to this problem, Hawkins and his colleagues proposed a

1364-6613/00/$ – see front matter © 2000 Elsevier Science Ltd. All rights reserved. PII: S1364-6613(00)01545-X
432
Trends in Cognitive Sciences – Vol. 4, No. 11, November 2000
 Luck et al. – ERP studies of attention Review

Box 1. Reaction time for the 21st century

(a) Stimulus 1 Stimulus 2 ... Stimulus N

EEG

(b)
Averaged ERP
Stimulus 1 –
N1
N2

Voltage (µV)
Stimulus 2
P1 P2

P3
Stimulus N +
0 200 400 600
Time (ms)
trends in Cognitive Sciences

Fig. I. Extraction of the ERP waveform from the ongoing EEG. (a) Stimuli (1… N) are presented while the EEG is being
recorded, but the specific response to each stimulus is too small to be seen in the much larger EEG. (b) To isolate the ERP from the
ongoing EEG, the EEG segments following each stimulus are extracted and averaged together to create the averaged ERP waveform.

In the early years of cognitive psychology, reaction-time meas- The ERP technique is relatively straightforward. ERPs begin
urements were incredibly useful for understanding a broad as postsynaptic potentials generated during neurotransmission.
range of cognitive processes, ranging from perception to mem- These electrical potentials passively travel through the brain and
ory, language, and motor programming. As we enter the 21st skull to the scalp, where they contribute to the overall electroen-
century, the techniques of cognitive neuroscience – especially cephalogram (EEG). The EEG is recorded from a set of elec-
ERPs and functional neuroimaging techniques – are beginning trodes on the surface of the scalp, and time-locked signal-averag-
to serve as high-tech substitutes for reaction-time measure- ing is used to extract the very small ERPs from the much larger
ments. There are three main reasons for this. First, and most ob- EEG. As shown in Fig. I, the segment of EEG following each
viously, they provide a link to the exploding field of neuro- stimulus (or each response) is extracted from the EEG, and these
science. Second, and less obviously, they are intrinsically segments are then lined up in time and averaged. Any brain
multidimensional measures of processing and are therefore well activity that is unrelated to the stimulus will average to zero
suited to separately measuring the subcomponents of cognition. (assuming a large number of trials), and any brain activity that is
For example, a single trial of a typical reaction-time experiment consistently time-locked to the stimulus will remain in the aver-
consists of a stimulus followed by a response, with no direct age. The resulting averaged ERP waveform consists of several
means of observing the processing that occurs between them. In positive and negative deflections that are called ‘peaks,’ ‘waves,’
an ERP experiment, in contrast, the stimulus elicits a continu- or ‘components,’ and these peaks are typically named with a P or
ous ERP waveform that can be used to directly observe neural N to indicate positive or negative and a number to indicate the
activity that is interposed between the stimulus and the re- timing of the peak (e.g. ‘P1’ for the first positive peak or ‘P110’
sponse. Moreover, the distribution of voltage over the scalp can to indicate a precise latency of 100 ms). The sequence of com-
be used to further isolate specific cognitive processes. Functional ponents following a stimulus reflects the sequence of neural
neuroimaging techniques do not provide a high-resolution meas- processes triggered by the stimulus, beginning with early sen-
ure of processing between a stimulus and a response, but they sory processes and proceeding through decision- and response-
instead provide a very detailed 3-D map of brain activity in related processes. The amplitude and latency of the successive
which processes can be isolated spatially. A third valuable aspect peaks can be used to measure the time course of cognitive pro-
of these techniques is that they allow processing to be measured cessing, and the distribution of voltage over the scalp can be used
in the absence of overt responses; in attention research, this is to estimate the neuroanatomical loci of these processes.
particularly important because it is useful to compare the pro-
cessing of attended and unattended stimuli without requiring Reference
subjects to respond to the unattended stimuli (for evidence that a Eimer, M. (1994) ‘Sensory gating’ as a mechanism for visuospatial
requiring subjects to respond to nominally unattended stimuli orienting: electrophysiological evidence from trial-by-trial cuing
changes the nature of attentional processing, see Ref. a). experiments. Percept. Psychophys. 55, 667–675

model in which there are two sources of noise, external stim- increase in sensory gain amplifies the stimulus noise but not
ulus noise and internal neural noise; an attention-related the neural noise, leading to improved perception13.

433
Trends in Cognitive Sciences – Vol. 4, No. 11, November 2000
Review Luck et al. – ERP studies of attention

Box 2. The time course of attention shifting

It is generally accepted that attention can shift from location to lo- with slightly different flicker frequencies at the two locations.
cation or from object to object, but there is significant controversy Each trial contained a cue stimulus that directed the subjects to
over the time course of these shifts, with some investigators arguing attend to one of these two locations in anticipation of a target at
that attention can shift approximately every 50 ms (Ref. a) and oth- the cued location. The task-irrelevant flickering was present in the
ers arguing that shifts of attention require about 500 ms (Ref. b). period between the cue and the target, and the ERP activity elic-
These arguments are based on inferences from behavioral data, but ited by the flickering was used to measure sensory responsiveness
it is difficult to directly observe shifts of attention from such data. at the two locations during this period. The ERP responses corre-
ERPs, in contrast, can provide an on-line, millisecond-by-milli- sponding to the two locations were isolated by taking advantage
second measure of the focus of attention, making it possible to of their slightly different flicker rates. These responses became
measure the timing of attention shifts somewhat more directly. enhanced 600–800 ms after the onset of the cue, indicating that
One recent ERP study measured the speed of attentional several hundred milliseconds were required to shift attention and
switching between objects during a difficult visual search task increase sensory responsiveness. This is in marked contrast with
(Ref. c). To assess the direction of attention at any given the rate of attention switching obtained in the visual search study
moment, this study examined the N2pc ERP component, described above, indicating that the time course of attention may
which is distinguished by being more negative over the cerebral differ greatly depending on the experimental paradigm.
hemisphere contralateral to the attended object (Ref. d). As
References
attention shifts from the left visual field to the right visual field, a Wolfe, J.M. (1994) Guided search 2.0: a revised model of visual
this ERP component shifts from the right hemisphere to the search. Psychonomic Bull. Rev. 1, 202–238
left hemisphere, providing a millisecond-by-millisecond meas- b Duncan, J. et al. (1994) Direct measurement of attentional dwell
ure of the shifting of attention. This study found that, even in time in human vision. Nature 369, 313–315
a very difficult visual search task, attention can shift serially c Woodman, G.F. and Luck, S.J. (1999) Electrophysiological
measurement of rapid shifts of attention during visual search.
from object to object approximately every 100 ms.
Nature 400, 867–869
Another study used ERPs to measure the time course of the
d Luck, S.J. and Hillyard, S.A. (1994) Spatial filtering during visual
shifting of attention following a symbolic attention-directing cue search: evidence from human electrophysiology. J. Exp. Psychol.
(Ref. e). To measure the direction of attention, a steady-state vis- Hum. Percept. Perform. 20, 1000–1014
ual evoked potential (SSVEP) paradigm was used in which task- e Müller, M.M. et al. (1998) The time course of cortical facilitation
irrelevant stimuli flickered rapidly at two possible target locations, during cued shifts of spatial attention. Nat. Neurosci. 1, 631–634

Lu and Dosher have recently proposed a more detailed ver- signal without amplifying the neural noise, leading to substan-
sion of this model and have provided a novel psychophysical tial improvements in perceptual accuracy. This is the pattern of
approach to distinguishing between gain control mechanisms results that was obtained, which provides psychophysical sup-
and other potential mechanisms of attention14. In their psy- port for the gain-control interpretation of the ERP findings.
chophysical experiments, subjects are presented with stimuli at However, this does not imply that attention always operates in
attended and ignored locations, and the stimuli are embedded this manner – as discussed below, attention may operate rather
in varying levels of visual noise. When the stimulus contains a differently in other experimental paradigms.
large amount of noise, accuracy is limited primarily by this
stimulus noise, and increasing the sensory gain at the attended Neural substrates of early selection
location should amplify both the signal and the noise, leading As described in Box 3, it is extremely difficult to localize the
to minimal improvements in perceptual accuracy. In contrast, neural generator sources of an ERP component. However,
when the stimulus contains very little noise, accuracy is limited significant progress has recently been made in identifying the
by neural noise rather than by stimulus noise, and increasing neural origins of the ERP attention effects, particularly the P1
the sensory gain at the attended location should amplify the modulation. First, it has been possible to demonstrate that an
earlier ERP component called the C1 wave is generated in
striate cortex (area V1; see Box 3)15,16 but is not influenced by
Onset of N1 Attend
attention right attention17–19. This finding indicates that attention operates
effect after information has passed through striate cortex. Second,
–1 µV N2 dipole modeling studies of the P1 wave have shown that its
distribution over the scalp is consistent with a neural genera-
–100 100 200 300 400
+ tor source in lateral extrastriate cortex17. Third, Wijers and his
colleagues have shown that both the P1 wave itself and the ef-
fect of spatial attention on the P1 are present, although
+1 µV slightly delayed, for stimuli that are presented on an isolumi-
P1 P2
nant background20. Because isoluminant stimuli primarily ac-
Attend
left trends in Cognitive Sciences tivate ventral-stream areas, this finding suggests that the P1 at-
tention effect is generated within the ventral pathway. Fourth,
Fig. 1. Paradigm for using ERPs to study attention. Stimulus display (left) and idealized
results (right). Subjects fixate a central cross and attend either to the left or right visual field.
studies that have combined ERP recordings with positron
Stimuli are then presented to the left and right visual fields in a rapid sequence. In this ex- emission tomography (PET) have indicated that the P1 wave
ample, the ERP elicited by a left visual field stimulus contains larger P1 and N1 components is associated with activation in the posterior fusiform gyrus21,22
when the stimulus is attended (‘Attend left’) than when it is ignored (‘Attend right’).
and/or dorsal occipital areas23. Together, these converging

434
Trends in Cognitive Sciences – Vol. 4, No. 11, November 2000
 Luck et al. – ERP studies of attention Review

Box 3. Localizing ERPs

In most cases, the cortical activation underlying an ERP compo- oriented 180° from each other in the calcarine sulcus; this leads
nent can be well approximated by a current dipole (a tiny posi- to the prediction that ERP activity generated in area V1 should
tive–negative pair of electrical charges). Computing the distribu- be opposite in polarity for upper- versus lower-field stimuli.
tion of voltage that would be observed on the surface for a known Consistent with this prediction, the C1 wave is negative for
dipole location and orientation is called the ‘forward problem’, upper-field stimuli and positive for lower-field stimuli; com-
and it is easy to solve with relatively straightforward mathematics. bined with its timing (∼60–80 ms), this finding provides good
To localize an ERP generator on the basis of the observed scalp evidence that the C1 wave is generated in area V1 (Refs a,b).
distribution is called the ‘inverse problem,’ and this is much more Other studies have also used previous anatomical and physiologi-
of a problem. It can be solved if there are only 1–2 dipoles, but it cal results as the basis for predictions about ERP components
becomes extremely difficult if the number of dipoles is unknown (Refs c,d). This approach can also be combined with mathemati-
and potentially large (i.e. more than about 4). The reason for this cal modeling and with parallel functional neuroimaging experi-
is that the inverse problem is ‘ill-posed,’ meaning that there is no ments to provide converging evidence about the neural sources
single solution; in fact, there are infinitely many patterns of dipoles of an ERP component (Ref. e). However, there is no approach
that could explain any given distribution of voltage on the scalp. that is currently guaranteed to correctly localize an ERP genera-
There are two general ways in which investigators have tor, and it is necessary to retain a healthy degree of skepticism.
attempted to circumvent the underdetermined nature of the
inverse problem. The first is to use mathematical constraints to References
reduce the number of possible solutions. For example, on the basis a Clark, V.P. et al. (1995) Identification of early visually evoked
of biophysical principles, it is reasonable to assume that ERPs are potential generators by retinotopic and topographic analyses.
generated only in the cortical gray matter and always have an ori- Hum. Brain Mapp. 2, 170–187
entation that is perpendicular to the cortical surface. By taking this b Mangun, G.R. et al. (1993) Electrocortical substrates of visual
constraint into account, it is possible to reduce (but not eliminate) selective attention. In Attention and Performance Vol. XIV (Meyer,

the problem of multiple solutions for a given scalp distribution. D. and Kornblum, S., eds), pp. 219–243, MIT Press
c Brunia, C.H.M. and Vingerhoets, A.J.J.M. (1980) CNV and EMG
A second approach has been to forego mathematical proce-
preceding a plantar flexion of the foot. Biol. Psychol. 11, 181–191
dures for localization and to instead make specific, testable pre-
d Luck, S.J. et al. (1997) Bridging the gap between monkey
dictions about the anatomical source of an ERP component on
neurophysiology and human perception: an ambiguity resolution
the basis of the known anatomy and physiology of specific brain theory of visual selective attention. Cognit. Psychol. 33, 64–87
regions. For example, many previous studies have shown that the e Luck, S.J. (1999) Direct and indirect integration of event-related
topographic mapping of area V1 is organized such that the upper potentials, functional magnetic resonance images, and single-unit
and lower visual fields are mapped onto regions of V1 that are recordings. Hum. Brain Mapp. 8, 15–120

results indicate that spatial attention modulates sensory attention did not reliably influence neural activity in striate
processing in extrastriate areas of the ventral visual pathway. cortex. This result is also consistent with the ERP studies
The experimental paradigm shown in Fig. 1 has also described above that found no effect of attention on the stri-
been modified for use with single-cell recordings in macaque ate-cortex C1 wave17–19. In contrast, recent functional mag-
monkeys24. When both the attended and ignored locations netic resonance imaging (fMRI) studies have demonstrated
are inside the receptive field of a given neuron in extrastriate increased striate-cortex activity in the hemisphere contralat-
area V4, that neuron will fire at a faster rate for attended- eral to the attended location26,27. There are two likely expla-
location stimuli than for unattended stimuli. Moreover, this nations for this discrepant result. First, it is possible that the
attention effect begins at 60 ms poststimulus, which is the tasks used in the fMRI experiments caused attention to
onset of stimulus-evoked activity in V4. This provides even operate differently than in the ERP and single-cell experi-
stronger and more specific evidence that attention modu- ments. Second, it is possible that the fMRI results do not re-
lates the feedforward transmission of information through flect a modulation of the initial feedforward sensory re-
the visual system. However, although the single-cell atten- sponse, but instead reflect a tonic activation or a feedback
tion effects are somewhat similar to the P1 ERP attention signal.
effects, it is not yet clear that the P1 effect arises in area V4. To explore these possibilities, Martinez and her col-
Specifically, the single-cell V4 effects were observed only leagues conducted both ERP and fMRI recordings from a set
when both the attended and ignored locations were inside of subjects who performed the same spatial attention task for
the receptive field of the cell being recorded, whereas P1 both recordings28. Even though the task was held constant
effects are typically observed with attended and ignored across recording techniques, discrepant striate-cortex results
locations that are too far apart to fall within a single V4 were again obtained. That is, the C1 component was not in-
receptive field. In addition, V4 neurons rarely respond to fluenced by attention in the ERP recordings, but increased
stimuli that fall more than 1° into the ipsilateral visual field, striate activity was observed contralateral to the attended lo-
whereas the P1 wave can be observed for ipsilateral stimuli cation in the fMRI recordings. Thus, the fMRI effects prob-
that are very far away from the midline. Thus, it is likely that ably do not reflect a modulation of the initial feedforward
the P1 effect arises in a somewhat more anterior visual area sensory response, as indexed by the C1 wave, but rather re-
with larger receptive fields, such as area TE. flect a tonic increase in neural activity that would not influ-
Recordings were also obtained from striate cortex in this ence the stimulus-triggered ERP waveform or some sort of
study, and as in the classic study of Moran and Desimone25, feedback signal that would influence ERP activity at a later

435
Trends in Cognitive Sciences – Vol. 4, No. 11, November 2000
Review Luck et al. – ERP studies of attention

time. In support of a feedback effect in V1, a recent single- surfaces42–45. Location is represented at the very earliest stages
unit study found that attention did not modulate the initial of the visual system, followed at later stages by surfaces and
V1 response, but instead influenced the V1 firing rate begin- then objects, and it therefore seems plausible that surface- and
ning approximately 200 ms after the initial response29. Thus, object-based attention effects might arise at a later stage than
an fMRI effect in area V1 does not imply a modulation of the spatial attention effects. However, a recent ERP study has un-
initial sensory response, and the lack of a C1 ERP effect does ambiguously demonstrated that attending to one of two
not imply that V1 activity is not affected by attention; to ob- superimposed surfaces yields the same pattern of enhanced
tain a complete picture, both types of evidence are needed. sensory processing that has been observed in studies of spatial
attention46. The surfaces were defined by patterns of red and
Early selection in other paradigms green dots that moved in such a manner that they yielded the
The ERP studies described so far have primarily used vari- perception of two transparent surfaces, one red and one green,
ants of the experimental paradigm shown in Fig. 1, but this sliding across each other. Even though the two perceived sur-
is not a paradigm that has been widely used in traditional faces were completely overlapping in space, larger P1 waves
cognitive studies of attention. However, similar results have were observed for a given surface when it was attended than
been observed with more common tasks such as spatial when the other surface was attended. Moreover, when the
cuing3,8,30 and visual search31,32. perception of two separate surfaces was eliminated by using
In spatial cuing studies, a cue stimulus directs the sub- stationary colored dots, the ERP modulations were elimi-
ject to attend to a particular location on each trial, and a tar- nated; this indicates that ERP modulations observed for the
get is then presented either at the attended location or at an moving stimuli reflected the allocation of attention to the
unattended location. Subjects are required to respond to the motion-and-color-defined surface rather than to the color
target no matter where it appears, and many studies have alone. Thus, directing attention to a surface (or to a surface-
found that responses are faster and more accurate when the defining motion pattern) appears to influence processing just
target appears at the cued location (valid trials) than when as early as directing attention to a region of 2D space.
the target appears at an uncued location (invalid trials)33,34.
Similarly, larger sensory-evoked ERP responses are observed Late selection in dual-task paradigms
for the targets on valid trials than on invalid trials3,8,30, indi- Selection occurs at early stages of processing under some
cating that the enhanced behavioral speed and accuracy are conditions, but there are many conditions under which
caused, at least in part, by enhanced sensory processing. both attended and unattended stimuli are fully identified
When the attention-directing cue is presented at the to- and attention operates at a post-perceptual stage. Lavie47,48
be-attended location (a peripheral cue) rather than at fix- has proposed that selection operates at an early stage only
ation (a central cue), attention may be summoned automat- under conditions of high perceptual load. This makes sense:
ically to the cued location. For example, even if the cue is why should attention suppress the perception of an unat-
nonpredictive such that the target is equally likely to appear tended stimulus unless the perceptual system is overloaded?
at the cued location or at an uncued location, peripheral cues Luck and Hillyard11 have extended this logic beyond the
will still cause an orienting of attention to the cued lo- simple early-late dichotomy, proposing that attention can
cation35. If the delay between a nonpredictive peripheral cue operate in a variety of cognitive subsystems (e.g. early sen-
and a target is short, responses are faster on valid trials than sory analysis, object recognition, working memory,
on invalid trials. However, if the delay between the cue and response selection, etc.) depending on the nature of the
the target is long, then responses are found to be slower on stimuli and the task, such that selective processing will
valid trials than on invalid trials. This phenomenon is called occur in a given subsystem when that subsystem suffers
inhibition of return, because it is thought to reflect a bias from interference due to the competing demands of multi-
against revisiting a location that has recently been attended36. ple stimuli or tasks. For example, if the task requires
Until recently, it was not known whether the excitatory subjects to discriminate the identity of a target embedded in
effect at short delays or the inhibitory effect at long delays are a dense array of similar distractors, then attention will be
entirely due to changes in motor response thresholds or to used to eliminate distractor interference during the percep-
changes in sensory responsiveness. However, recent psy- tion of the target. In contrast, if subjects are presented with
chophysical studies have shown that both the excitatory and a stream of 30 colored letters at a rate of one letter per
inhibitory effects are at least partially due to changes in per- 500 ms and are required to report all of the red letters at the
ceptual quality, with improved perception on valid trials at end of the trial, then there will be no need to suppress the
short delays and impaired perception on valid trials at long non-red letters before they are perceived, but it will be
delays37,38. Similarly, ERP studies have shown that the sen- necessary to store only the red letters in working memory to
sory-evoked P1 wave is enhanced on valid trials at short delays avoid exceeding the limited capacity of working memory.
and suppressed on valid trials at long delays39–41. Thus, both
the excitatory and the inhibitory effects of peripheral cues re- Isolating different cognitive subsystems
flect, at least in part, modulations of sensory processing. Recent ERP studies have supported this conceptualization of
Many recent cognitive studies of attention have also fo- attention by showing that attention operates in different cogni-
cused on the issue of whether attention is directed to spatial tive subsystems for different tasks. In particular, the early-selec-
locations per se or to the objects at those locations. These stud- tion studies described in the preceding sections have been
ies have indicated that, depending on the conditions, complemented by ERP studies of two paradigms in which at-
attention can be directed to locations, to objects, or to tention would be expected to operate at later stages, namely the

436
Trends in Cognitive Sciences – Vol. 4, No. 11, November 2000
 Luck et al. – ERP studies of attention Review

(a) (b)
T1 R1
T1 T2 R1 R2 Task 1 RT1
Task 2 SOA
EFHJVTOGXPHCN T2 RT2 R2
Time 800 Time
100
T2 accuracy (% correct)

T2 reaction time (ms)

80 700

60 600
40
500
20

0 400
0 1 2 3 4 5 6 7 8 0 100 200 300 400
T1–T2 lag T1–T2 SOA (ms)
(c) +6 (d)
P3
800
Component amplitude (µV)

+4
RT

700
+2
Latency (ms)

P1
600
0
P3
N1 500
–2

N400
–4 400
Lag 1 Lag 3 Lag 7 0 100 200
300 400
T1–T2 SOA (ms)
trends in Cognitive Sciences
Fig. 2. Attentional paradigms. (a) Typical attentional blink experiment64. Stimuli are presented at a rate of about 10 per second at fix-
ation. The targets might be white letters presented in a stream of black non-targets, or the presence of a particular letter. In this case, at the
end of each 2 s trial, the subject reports the identity of the one white letter (T1) and the presence or absence of the letter X (T2). The lag be-
tween T1 and T2 varies across trials, and accuracy for detecting T2 is found to be highly impaired at lags 2–5 (corresponding to 200–500 ms).
(b) Schematic psychological refractory period experiment. On each trial, two highly discriminable targets (T1 and T2) are presented without
any distractors, and subjects make speeded responses to both targets. When the stimulus onset asynchrony (SOA) between T1 and T2 is long,
both tasks can be performed quickly (response times, RT1 and RT2, are similar). However, when the SOA is short, the response to T2 becomes
delayed. (c) Summary of several ERP experiments examining different ERP components in attentional blink experiments. No suppression was
observed at lag 3 for the P1, N1, or N400 components, but the P3 wave was completely eliminated at lag 3. (Adapted from Ref. 51.) (d) P3
latency and reaction time in a psychological refractory period experiment. Although reaction time was approximately 200 ms longer at the
short SOA than at the long SOA, P3 latency was only slightly longer at the long SOA. (Adapted from Ref. 59.)

attentional blink paradigm and the psychological refractory pe- about 20 stimuli presented at a rate of about 10 stimuli per
riod paradigm. These paradigms are very different from those second, and the subject is required to report the two targets
used to demonstrate early selection, but this is sensible given at the end of the trial. The psychological refractory period is
that different paradigms will stress different cognitive subsys- similar, but has two main differences. First, the targets are
tems. Ultimately, it would be useful to show that the locus of presented in isolation, without any distractors. Second, sub-
attention shifts systematically from stage to stage as a result of jects are asked to respond to each target as quickly as poss-
parametric manipulations in a single experimental paradigm, ible rather than waiting until the end of the trial to respond.
but such an experiment has not yet been published. However, In both paradigms, it is reasonable to suppose that it is
the existing studies are sufficient to demonstrate that attention difficult to process T2 while T1 is being processed, and this
operates at different stages under different conditions. is borne out by the typical results: When the amount of
The attentional blink and psychological refractory pe- time between T1 and T2 is relatively short, responses to T2
riod paradigms are illustrated in Fig. 2a,b. Both paradigms are less accurate (in the attentional blink paradigm) or
involve the presentation of two targets – labeled T1 and slower (in the psychological refractory period paradigm).
T2 – on each trial. In the attentional blink paradigm, these When the amount of time between T1 and T2 is suffi-
targets are embedded in a rapid stream of non-targets, all ciently long, however, the processing of T1 is largely com-
presented visually at fixation. Each trial typically consists of plete before T2 is presented, leading to a return to baseline

437
Trends in Cognitive Sciences – Vol. 4, No. 11, November 2000
Review Luck et al. – ERP studies of attention

Box 4. Isolating overlapping ERP waveforms

Context word T1 T2 which T2 was expected to elicit a
small N400 from trials on which
T2 was expected to elicit a large

WPSCDSN
DPWVCPB
GRSDPKN

WDPTBNF

RMVCPKL
PNVCSZP

CBNDPNJ
DLJJCNW
SPLDJMF

FDLNLKB
N400. N400 amplitude was

3333333
RAZOR

SHAVE
modulated by a manipulation of
(a) semantic mismatch: previous
studies have shown that words
that mismatch the current
(b) Matching semantic context generate a large
T2 trial N400 (e.g. the last word in the
Additional activity due sentence, ‘The monkey eagerly
to semantic mismatch bit into the ripe, yellow tele-
peaking 400 ms after T2 phone’), and words that match
(c) Mismatching the semantic context produce lit-
T2 trial tle or no N400 (e.g. the last
word in the sentence, ‘The mon-
T2-related N400 key eagerly bit into the ripe, yel-
Mismatching low banana’). In our experiment,
(d) minus we established a semantic con-
matching Time text at the beginning of each trial
trends in Cognitive Sciences
by presenting a ‘context word’
Fig. I. Overlapping ERP components during presentation of a rapid stream of stimuli.
that subjects were required to
(a) The stimuli on a typical trial, beginning with a 1 s presentation of a context word followed
remember. The first target (T1)
by a stream of stimuli presented at a rate of 10 per second (all stimuli were presented in nor-
mal upright orientation at fixation). The targets were a number (T1) followed by a word (T2). was a number, and the second
At the end of each trial, subjects reported whether T1 was odd or even and whether T2 was (T2) was a word, either semanti-
semantically related or semantically unrelated to the context word. (b) The overlapping ERP cally related or semantically
components for a trial in which the T2 word matches the semantic context for that trial. (c) The unrelated to the context word.
activity for a semantic mismatch trial, which is equivalent to the matching trials plus the addi- As shown in Fig. I, match and
tion of mismatch-related ERP activity following T2. (d) The result of subtracting the matching
mismatch trials will have largely
trials from the mismatching trials, which isolates mismatch-related activity triggered by T2
(primarily the N400 component).
identical patterns of ERP overlap
from the preceding and subse-
The use of ERP recordings in tasks with rapid streams of quent items in the stream. The only difference between these tri-
stimuli leads to significant technical challenges. Specifically, als will be that the semantic mismatch trials should elicit a larger
each item presented in a rapid stream produces an ERP re- N400 than the semantic match trials. The mismatch-minus-
sponse that lasts for several hundred milliseconds, long past the match subtraction therefore isolates the T2-related N400
onset of the next stimulus. Consequently, the ERP elicited by a response. The manipulation of semantic mismatch was factor-
given item will be overlapped by the ERPs elicited by previous ially crossed with a manipulation of T1–T2 lag so that the N400
and subsequent items, making it difficult to isolate the ERP for could be measured before, during, and after the attentional blink
each individual stimulus. Under certain conditions, this overlap period. Similar approaches have been used to isolate the P3 wave
can be eliminated mathematically (Ref. a), but in many cases a (Ref. c) and the lateralized readiness potential (Ref. d) in studies
simple subtraction procedure can be used both to eliminate of the psychological refractory period.
overlap and to isolate specific ERP components.
The essence of this approach is to manipulate an experi- References
mental variable that is known to influence the amplitude of a a Woldorff, M. (1993) Distortion of ERP averages due to overlap from

particular ERP component and to subtract the ERP waveform temporally adjacent ERPs: analysis and correction. Psychophysiology
30, 98–119
from trials on which this ERP component is small from the
b Luck, S.J. et al. (1996) Word meanings can be accessed but not
ERP waveform from trials on which it is large. This manipu-
reported during the attentional blink. Nature 382, 616–618
lation is then factorially crossed with the experimental manip-
c Luck, S.J. (1998) Sources of dual-task interference: evidence from
ulation of interest. For example, we have used this approach in human electrophysiology. Psychol. Sci. 9, 223–227
an attentional blink experiment to isolate the N400 elicited by d Osman, A. and Moore, C.M. (1993) The locus of dual-task interference:
the second target (T2) (Ref. b). As illustrated in Fig. I, we iso- psychological refractory effects on movement-related brain
lated the T2-related N400 component by subtracting trials on potentials. J. Exp. Psychol. Hum. Percept. Perform. 19, 1292–1312

for both reaction time and accuracy. There is one notable Although these paradigms are conceptually similar and
exception to the similarity in the results across these para- lead to largely parallel results, they stress different cognitive
digms, namely that accuracy in the attentional blink para- subsystems. Specifically, the use of speeded responses in the
digm is not usually impaired when T2 immediately follows psychological refractory period creates the potential for inter-
T1, whereas reaction times increase monotonically as the ference between T1 processing and T2 processing at the stages
T1–T2 delay decreases in the psychological refractory of response selection and execution, but the stimuli are so sim-
period paradigm (the reasons for this are complex – for a ple that there should be little or no interference at earlier
detailed discussion, see Ref. 49). stages. In contrast, responses are unspeeded in the attentional

438
Trends in Cognitive Sciences – Vol. 4, No. 11, November 2000
 Luck et al. – ERP studies of attention Review

blink paradigm, minimizing response-related interference.

Outstanding questions
Moreover, the stimuli are still fairly simple, and it should be
simple to identify every stimulus in the stream even at a rate of • It is now clear that attention can operate in different cognitive subsystems
10 stimuli per second50. The likely source of interference in under different conditions, but it is not yet clear whether there are
this paradigm, then, is the possibility that the representation of completely different attention mechanisms operating in these different
one stimulus might be overwritten by the representation of subsystems. Do these different attention effects reflect (a) completely
independent attention mechanisms? (b) a collection of loosely
the next stimulus unless the relevant stimuli are stored in a interconnected but somewhat independent attention mechanisms? (c) a
durable form. That is, previous research has indicated that it set of separable attention mechanisms that are centrally controlled? or (d)
may take several hundred milliseconds to store a visual stimu- a unitary attention mechanism that can operate in a variety of subsystems?
lus in working memory such that it cannot be overwritten by • Studies that have combined PET and ERP measures have found an area in
subsequent stimuli, and attention may be necessary to prevent the posterior fusiform gyrus that may be the generator source of the P1
spatial attention effect21,22, but another study indicated that the
the stimuli following T1 from interfering with the process of generator might instead be in the dorsal occipital cortex23. What is the
storing T1 in working memory. Consequently, if T2 is actual generator site, and how can we be sure?
presented shortly after T1, it will fail to be stored in working • Both ERP and psychophysical studies have been used to estimate the
memory and will be unavailable for report at the end of speed with which attention can shift from one location to the next, but
the trial. these estimates range from less than 50 ms per shift to over 500 ms per
shift (see Box 2). What is responsible for these radically different
estimates of the time course of attentional orienting?
ERP support for a multiple subsystems view • ERP studies suggest that the impairment in T2 processing arises from
This theoretical analysis of the attentional blink and psycho- different stages of processing in the attentional blink and psychological
logical refractory period paradigms has been supported by re- refractory period paradigms51,52,59. However, recent behavioral studies
cent ERP experiments (see Box 4 for a discussion of the meth- indicate that these impairments reflect a single attentional
bottleneck62,63. How can this discrepancy be resolved?
ods necessary for isolating ERPs under conditions of rapid
stimulation). In the attentional blink paradigm, a variety of
ERP components have been compared to determine the first resolution of the ERP technique should make finer distinctions
stage at which processing is suppressed during the attentional possible, especially when appropriate experimental designs are
blink51,52. As illustrated in Fig. 2c, no change in amplitude or used to highlight specific cognitive subsystems. In the future,
latency was observed for the P1, N1, or N400 components, the ability to isolate attentional processes to specific cognitive
which respectively reflect early sensory analysis53, visual dis- subsystems should make it possible to replace blanket state-
crimination processes54, and semantic analysis55. However, the ments about attention (e.g. ‘Attention is not needed for the de-
P3 wave – which is thought to reflect the updating of working tection of simple features’ or ‘Attention is distributed like a gra-
memory56 – was completely suppressed during the attentional dient’) with much more specific statements about how
blink. This is consistent with the proposal that the attentional attention operates within a particular subsystem (e.g. ‘Attention
blink occurs after perception is complete and reflects a failure binds features together to facilitate visual form recognition’).
to store T2 in working memory while T1 is being analyzed57,58.
A very different pattern of results has been obtained in
Acknowledgements
psychological refractory period experiments. In these experi-
Preparation of this article was supported by grant SBR 98-09126 from the
ments59, the P3 wave is slightly reduced in amplitude and National Science Foundation, by grant MH56877 from the National
slightly delayed in latency when the interval between T1 and Institute of Mental Health, and by grant RG0136 from the Human
T2 is shortened, but these effects are not nearly large enough Frontier Science Program.
to account for the large delays in reaction time (see Fig. 2d).
However, the lateralized readiness potential – which reflects
References
motor preparation60 – is significantly delayed when the in-
1 Eason, R. et al. (1969) Effects of attention and arousal on visually
terval between T1 and T2 is shortened61. Thus, attention- evoked cortical potentials and reaction time in man. Physiol. Behav. 4,
related modulations appear to occur primarily in response- 283–289
related processes in the psychological refractory period 2 Duncan, J. (1980) The locus of interference in the perception of
paradigm, whereas the modulation occurs in working mem- simultaneous stimuli. Psychol. Rev. 87, 272–300
3 Luck, S.J. et al. (1994) Effects of spatial cuing on luminance
ory in the attentional blink paradigm. That is, response-
detectability: psychophysical and electrophysiological evidence for
related processing is modulated by attention in a task that early selection. J. Exp. Psychol. Hum. Percept. Perform. 20, 887–904
stresses response systems by requiring rapid responses, 4 Hillyard, S.A. et al. (1973) Electrical signs of selective attention in the
whereas working memory is modulated by attention in a human brain. Science 182, 177–179
paradigm that stresses working memory by requiring multi- 5 Hillyard, S.A and Münte, T.F. (1984) Selective attention to color and
location: an analysis with event-related brain potentials. Percept.
ple items to be stored in a short period of time.
Psychophys. 36, 185–198
6 Mangun, G.R. and Hillyard, S.A. (1988) Spatial gradients of visual
Conclusions attention: behavioral and electrophysiological evidence.
Together with the experiments on early selection described at Electroencephalogr. Clin. Neurophysiol. 70, 417–428
the beginning of this article, these experiments illustrate how 7 Neville, H.J. and Lawson, D. (1987) Attention to central and peripheral
visual space in a movement detection task. I. Normal hearing adults.
ERPs have been used to show that attention can operate in dif-
Brain Res. 405, 253–267
ferent cognitive subsystems under different conditions. Most 8 Eimer, M. (1994) ‘Sensory gating’ as a mechanism for visuospatial
existing ERP studies have explored coarsely defined cognitive orienting: electrophysiological evidence from trial-by-trial cuing
subsystems (e.g. early versus late), but the excellent temporal experiments. Percept. Psychophys. 55, 667–675

439
Trends in Cognitive Sciences – Vol. 4, No. 11, November 2000
Review Luck et al. – ERP studies of attention

9 Rugg, M.D. et al. (1987) Modulation of visual event-related potentials 36 Posner, M.I. and Cohen, Y. (1984) Components of visual orienting. In
by spatial and non-spatial visual selective attention. Neuropsychologia Attention and Performance Vol. X (Bouma, H. and Bouwhuis, D.G.,
25, 85–96 eds), pp. 531–556, Erlbaum
10 Robinson, D.L. and Rugg, M.D. (1988) Latencies of visually responsive 37 Handy, T.C. et al. (1999) Promoting novelty in vision: inhibition of
neurons in various regions of the Rhesus monkey brain and their return modulates perceptual-level processing. Psychol. Sci. 10, 157–161
relation to human visual responses. Biol. Psychol. 26, 111–116 38 Luck, S.J. and Thomas, S.J. (1999) What variety of attention is
11 Luck, S.J. and Hillyard, S.A. (1999) The operation of selective attention automatically captured by peripheral cues? Percept. Psychophys.
at multiple stages of processing: evidence from human and monkey 61, 1424–1435
electrophysiology. In The New Cognitive Neurosciences (2nd edn) 39 Hopfinger, J.B. and Magnun, G.R. (1998) Reflective attention
(Gazzaniga, M.S., ed.), pp. 687–700, MIT Press modulates processing of visual stimuli in human extrastriate cortex.
12 Hillyard, S.A. et al. (1998) Sensory gain control (amplification) as a Psychol. Sci. 9, 441–447
mechanism of selective attention: electrophysiological and 40 McDonald, J.J. et al. (1999) An event-related brain potential study of
neuroimaging evidence. Philos. Trans. R. Soc. London 353, 1257–1270 inhibition of return. Percept. Psychophys. 61, 1411–1423
13 Hawkins, H.L. et al. (1990) Visual attention modulates signal 41 Eimer, M. (1994) An ERP study on visual spatial priming with peripheral
detectability. J. Exp. Psychol. Hum. Percept. Perform. 16, 802–811 onsets. Psychophysiology 31, 154–163
14 Lu, Z. and Dosher, B.A. (1998) External noise distinguishes attention 42 Duncan, J. (1984) Selective attention and the organization of visual
mechanisms. Vision Res. 38, 1183–1198 information. J. Exp. Psychol. Gen. 113, 501–517
15 Clark, V.P. et al. (1995) Identification of early visually evoked potential 43 Egly, R. et al. (1994) Shifting visual attention between objects and
generators by retinotopic and topographic analyses. Hum. Brain locations: evidence from normal and parietal lesion subjects. J. Exp.
Mapp. 2, 170–187 Psychol. Gen. 123, 161–177
16 Jeffreys, D.A. and Axford, J.G. (1972) Source locations of pattern- 44 Vecera, S.P. and Farah, M.J. (1994) Does visual attention select objects
specific components of human visual evoked potentials. I: components or locations? J. Exp. Psychol. Gen. 123, 146–160
of striate cortical origin. Exp. Brain Res. 16, 1–21 45 He, Z.J. and Nakayama, K. (1992) Surfaces versus features in visual
17 Clark, V.P. and Hillyard, S.A. (1996) Spatial selective attention affects search. Nature 359, 231–233
early extrastriate but not striate components of the visual evoked 46 Valdes-Sosa, M. et al. (1998) Switching attention without shifting the
potential. J. Cogn. Neurosci. 8, 387–402 spotlight object-based attentional modulation of brain potentials.
18 Mangun, G.R. et al. (1993) Electrocortical substrates of visual selective J. Cogn. Neurosci. 10, 137–151
attention. In Attention and Performance Vol. XIV (Meyer, D. and 47 Lavie, N. and Tsal, Y. (1994) Perceptual load as a major determinant of the
Kornblum, S., eds), pp. 219–243, MIT Press locus of selection in visual attention. Percept. Psychophys. 56, 183–197
19 Gomez Gonzales, C.M. et al. (1994) Sources of attention-sensitive 48 Lavie, N. (1995) Perceptual load as a necessary condition for selective
visual event-related potentials. Brain Topogr. 7, 41–51 attention. J. Exp. Psychol. Hum. Percept. Perform. 21, 451–468
20 Wijers, A.A. et al. (1997) An ERP study of visual spatial attention and 49 Visser, T.A.W. et al. (1999) Attentional switching in spatial and
letter target detection for isoluminant and nonisoluminant stimuli. nonspatial domains: evidence from the attentional blink. Psychol. Bull.
Psychophysiology 34, 553–565 125, 458–469
21 Heinze, H.J. et al. (1994) Combined spatial and temporal imaging of 50 Potter, M.C. (1976) Short-term conceptual memory for pictures. J. Exp.
brain activity during visual selective attention in humans. Nature Psychol. Hum. Percept. Perform. 2, 509–522
372, 543–546 51 Vogel, E.K. et al. (1998) Electrophysiological evidence for a
22 Mangun, G.R. et al. (1997) Covariations in ERP and PET measures of postperceptual locus of suppression during the attentional blink.
spatial selective attention in human extrastriate visual cortex. Hum. J. Exp. Psychol. Hum. Percept. Perform. 24, 1656–1674
Brain Mapp. 5, 273–279 52 Luck, S.J. et al. (1996) Word meanings can be accessed but not
23 Woldorff, M.G. et al. (1997) Retinotopic organization of early visual reported during the attentional blink. Nature 382, 616–618
spatial attention effects as revealed by PET and ERPs. Hum. Brain 53 Hillyard, S.A. and Picton, T.W. (1987) Electrophysiology of cognition. In
Mapp. 5, 280–286 Handbook of Physiology: Section 1. The Nervous System (Vol. 5, Part 2)
24 Luck, S.J. et al. (1997) Neural mechanisms of spatial selective attention (Plum, F., ed.), pp. 519–584, Waverly Press
in areas V1, V2, and V4 of macaque visual cortex. J. Neurophysiol. 54 Vogel, E.K. and Luck, S.J. (2000) The visual N1 component as an index
77, 24–42 of a discrimination process. Psychophysiology 37, 190–123
25 Moran, J. and Desimone, R. (1985) Selective attention gates visual 55 Kutas, M. (1997) Views on how the electrical activity that the brain
processing in the extrastriate cortex. Science 229, 782–784 generates reflects the functions of different language structures.
26 Somers, D.C. et al. (1999) Functional MRI reveals spatially specific Psychophysiology 34, 383–398
attentional modulation in human primary visual cortex. Proc. Natl. 56 Donchin, E. (1981) Surprise!…Surprise? Psychophysiology 18, 493–513
Acad. Sci. U. S. A. 96, 1663–1668 57 Chun, M.M. and Potter, M.C. (1995) A two-stage model for multiple
27 Gandhi, S.P. et al. (1999) Spatial attention affects brain activity in target detection in rapid serial visual presentation. J. Exp. Psychol.
human primary visual cortex. Proc. Natl. Acad. Sci. U. S. A. 96, 3314–3319 Hum. Percept. Perform. 21, 109–127
28 Martinez, A. et al. (1999) Involvement of striate and extrastriate visual 58 Shapiro, K.L. et al. (1997) The attentional blink: a view on attention
cortical areas in spatial attention. Nat. Neurosci. 2, 364–369 and a glimpse on consciousness. Trends Cognit. Sci. 1, 291–296
29 Roelfsema, P.R. et al. (1998) Object-based attention in the primary 59 Luck, S.J. (1998) Sources of dual-task interference: evidence from
visual cortex of the macaque monkey. Nature 395, 376–381 human electrophysiology. Psychol. Sci. 9, 223–227
30 Mangun, G.R. and Hillyard, S.A. (1991) Modulations of sensory- 60 Coles, M.G.H. et al. (1995) Mental chronometry and the study of human
evoked brain potentials indicate changes in perceptual processing information processing. In Electrophysiology of Mind: Event-Related
during visual-spatial priming. J. Exp. Psychol. Hum. Percept. Perform. Brain Potentials and Cognition. (Rugg, M.D. and Coles, M.G.H., eds)
17, 1057–1074 pp. 86–131, Oxford University Press
31 Luck, S.J. et al. (1993) Attention-related modulation of sensory-evoked 61 Osman, A. and Moore, C.M. (1993) The locus of dual-task interference:
brain activity in a visual search task. J. Cogn. Neurosci. 5, 188–195 psychological refractory effects on movement-related brain potentials.
32 Luck, S.J. and Hillyard, S.A. (1995) The role of attention in feature J. Exp. Psychol. Hum. Percept. Perform. 19, 1292–1312
detection and conjunction discrimination: an electrophysiological 62 Jolicoeur, P. (1998) Modulation of the attentional blink by on-line
analysis. Int. J. Neurosci. 80, 281–297 response selection: evidence from speeded and unspeeded Task-sub-1
33 Posner, M.I. et al. (1980) Attention and the detection of signals. J. Exp. decisions. Mem. Cognit. 26, 1014–1032
Psychol. Gen. 109, 160–174 63 Jolicoeur, P. (1999) Restricted attentional capacity between sensory
34 Posner, M.I. (1980) Orienting of attention. Q. J. Exp. Psychol. 32, 3–25 modalities. Psychonomic Bull. Rev. 6, 87–92
35 Jonides, J. (1981) Voluntary versus automatic control over the mind’s 64 Raymond, J.E. et al. (1992) Temporary suppression of visual processing
eye’s movement. In Attention and Performance Vol. IX (Long, J.B. and in an RSVP task: an attentional blink? J. Exp. Psychol. Hum. Percept.
Baddeley, A.D., eds), pp. 187–203, Erlbaum Perform. 18, 849–860

440
Trends in Cognitive Sciences – Vol. 4, No. 11, November 2000