Introduction

Ehlers and Clark (2000) present a cognitive model of posttraumatic stress disorder (PTSD) that
provides a framework for the cognitive-behavioral treatment of PTSD. The model proposes that two key
processes lead to a sense of current threat. The first process consists of negative appraisals of the trauma and/or
its sequel. A prospective longitudinal study found that appraisals of intrusive recollections contribute
significantly to the maintenance of PTSD (Ehlers, Mayou, & Bryant, 1998). Second, the nature of the trauma
memory and its link to other autobiographical memories can lead to persistent PTSD. According to Foa and
Rothbaum (1998) a pathological trauma memory structure consists of: (1) excessive response elements, (2)
erroneous stimuli-stimuli associations, (3) erroneous associations between harmless stimuli and the meaning
”dangerous”, and (4) erroneous associations that lead to the evaluation of oneself as incompetent. Furthermore,
Ehlers and Clark (2000) proposed that the trauma memory is poorly elaborated and inadequately integrated into
its context in time, place, subsequent and previous information and other autobiographical memories. This
hypothesis has been partially supported by studies in which individuals with PTSD displayed difficulties in
retrieving specific positive autobiographical memories (McNally, Lasko, Macklin; & Pitman, 1995; McNally,
Litz, Prassas, Shin, & Weathers, 1994).
Negative appraisals of intrusive recollections can be conceptualized as meaning elements in a
pathological trauma memory structure. Meaning elements consisting of danger interpretations of harmless
stimuli/responses can activate stimuli (intrusions) and excessive response elements (e.g. fear, avoidance) in a
trauma memory network. Other types of appraisals that can lead to an activation of a trauma network are (1)
negative appraisals of how oneself reacted during the trauma and its sequel, and (2) negative appraisals of other
peoples reactions during the trauma sequel (Dunmore, Clark, & Ehlers, 1999). If such activation occurs the
individual with PTSD ought to especially attend to the trauma-related stimuli and responses. A more elaborated
scanning on trauma-related stimuli and responses ought to lead to a memory bias for such material.
The trauma memory elements are internally cohesive and unintegrated with pre-trauma schematic
memory elements (Power, & Dalgleish, 1997). If individuals with PTSD have a memory bias for trauma-related
material, the hypothesized maintaining factor of the disorder is the underlying incomplete processing of the
elements in the trauma network that are unintegrated with pre-traumatic memory elements. A memory bias for
trauma-related stimuli/responses is only an index of such incomplete processing. Thus, one reason why PTSD is
maintained is because PTSD Ss do not have access to information from the pre-trauma memory network that is
incompatible to the trauma-related meaning elements. Also, such persons have difficulties in remembering
aspects of the trauma and its sequel that is at odds with their negative appraisals. As Ehlers and Clark (2000)
state, their recall is biased by their appraisals and they selectively retrieve information that is in accord with these
appraisals (pp. 326-327). As a consequence, individuals with PTSD ought to have inferior memory for non-
traumatic material.
A scarce amount of research has been carried out in order to test hypotheses about memory biases in
PTSD (Litz, Weathers, Monaco, Herman, Wulfsohn, Marx, & Keane, 1996; McNally, Metzger, Lasko, Clancy,
& Pitman, 1998; Vrana, Roodman, & Beckham, 1995). Litz et al. (1996) found a memory bias for trauma words
in PTSD Ss in comparison to a healthy control group, but not relative to a control group with other axis-I
disorders. Vrana et al. (1995) found no trauma-specific recall in PTSD Ss in comparison to controls. The bias
that was found in McNally et al.’s (1998) study stemmed from recall of fewer non-threat-related words by
individuals with PTSD. There are at least three possible explanations to the inconsistent results on memory
biases for trauma-related material in PTSD. Firstly, the type of independent stimuli that were used (single words)
may not have been meaningful enough in order to fully activate the trauma memory network in the PTSD Ss.
Secondly, the participants did not conduct emotionality ratings of the independent stimuli themselves. Thus, the
personal emotional significance of the independent stimuli for the Ss were unknown. Thirdly, this kind of
research is characterized by mainly using external stimuli (mostly words) as independent variables, and
compares different categories of external stimuli (e.g. trauma-specific words vs. trauma-general and emotional
words) in terms of the participant’s memory for these. According to Beck (1976), however, what is assumed to
lead to psychopathology is not external stimuli as such, but the way stimuli are interpreted. In this model,
emotional responses are caused not by external stimuli, but by the way these stimuli are interpreted, categorized,
and “encoded”. If the same applies also to people’s memory functioning, we should expect memory biases to
appear as a result of various kinds of interpretations, categorizations, or encodings of external stimuli, rather than
in relation to external stimuli as such.

The face recognition paradigm

The face recognition paradigm that was introduced by Lundh and Öst (1996a; 1996b) is an experimental
paradigm where the participants’ interpretations/categorizations of external stimuli are used as the independent
variable, and recognition performance is used as the dependent variable. By means of this paradigm, Lundh and
Öst (1996b) found that patients with social phobia showed evidence of a memory bias for faces that were
perceived as “critical” vs. faces that were perceived as “accepting”. 1 In another study of patients with social
phobia and matched controls (Lundh & Öst, 1996a), the participants were required to make another kind of
categorization, i.e., to rate whether the persons on the photos were individuals that they would probably get good
contact with or not, if they were to meet them in real life. Again, for each subject, the target faces were divided
into categories, depending on the subject’s ratings of the expected quality of contact. The results of this study
showed no significant differences between the social phobics and the controls. That is, these two studies (Lundh
& Öst, 1996a, 1996b) gave a first indication that if the participants’ interpretations/categorizations are used as
the independent variable, rather than categories of external stimuli, then we may find evidence of memory biases
of that kind that may be expected from the nature of the participants’ fears.
Further evidence of the validity of the face recognition paradigm was found in a second pair of studies by
Lundh, Thulin, Czyzykow, and Öst (1998). If the face recognition paradigm is sensitive to patients’ basic
emotional concerns, and specifically sensitive to the concerns that characterize each diagnostic group, then the
memory bias for critical vs. accepting faces that was found in social phobia should not be found in panic disorder
with agoraphobia. This was tested by Lundh et al. (1998, Study 1). Since the typical fear of a panic patient with
agoraphobia is not about being negatively evaluated by others, no bias for critical faces was expected, and no
such bias was found. Patients with this kind of diagnosis, on the other hand, are known to fear being in places or
situations where help may not be available in the event of having a panic attack or panic-like symptoms. It is a
basic concern of agoraphobics that they should have safe persons around, whom they can trust for help if they
should suffer a panic attack. To the extent that this produces a heightened attention to signs of whether other
people are safe persons or not, this might be expected to lead to a more elaborated encoding of faces which are
perceived as safe, and therefore to better memory for such faces. A further study was, therefore, designed (Lundh
et al., 1998, Study 2) where the participants were asked to judge whether they perceived the persons on the
photos as a safe or unsafe persons, i.e., “as a person who you would trust if you were in need of help, or as a
person who you would not trust in such a situation”.2 The results showed that the panic patients recognized more
safe than neutral faces whereas the controls did not so.
An essential thing about the face recognition paradigm is that the photos of human faces that are used as
stimuli are not chosen because of their physical characteristics, or their emotional expressions. 3 The photos that
are used as stimuli are ordinary photos of people that are probably relatively representative of how people look
in ordinary everyday situations (at least in front of the camera) with no pronounced facial emotional expressions.
The above-mentioned studies with social phobics (Lundh & Öst, 1996a, 1996b) and panic disorder with
agoraphobia (Lundh et al., 1998) used exactly the same external stimuli, i.e., exactly the same photos; the studies
only differed in terms of the participants’ encoding activities, i.e., their way of interpreting and categorizing the
faces on the photos. The purpose of the present study was to expand this experimental paradigm also to another
diagnostic group, i.e., patients with PTSD. According to Foa and Kozak (1986) anxiety disorders can be
conceptualized in terms of fear structures, or associative networks of fear that include stimuli, response and
meaning elements. From this perspective, the results with the face recognition paradigm can be described in
terms of whether the encoding tasks involved interpretive information that is central to the fear structure, and
that would thereby activate this structure and lead to a more extensive scan of faces when these were interpreted
as fear-relevant.4

1
The patients and a sex-and age-matched control group were shown twenty ordinary photos of persons (ten females, ten males) that were
taken in an everyday situation, and that were not arranged to depict any particular facial expressions. As encoding task, the participants were
required to judge whether they perceived the person on the photo as a critical or accepting person, i.e., “as a person who is generally critical
towards others and tends to find faults with other people, or as generally accepting and tolerant with regard to others’ shortcomings”. This
rating was done on a 5-point scale, where 1 meant ”a very accepting person”, 2 ”a fairly accepting person”, 3 ”a person who is neither
accepting nor critical”, 4 ”a fairly critical person”, and 5 ”a very critical person”. For each subject, the target faces were then divided into
three categories, depending on the subject’s rating of the degree of critical or accepting attitudes of the persons seen: Faces rated as 4 or 5
were categorized as critical, faces rated as 3 were categorized as neutral, and faces rated as 2 or 1 were categorized as accepting. The results
showed that the social phobics recognized significantly more of the “critical” faces than the “accepting” faces, whereas the controls had a
tendency to an opposite bias for the “accepting” faces.
2
Again, this rating was done on a 5-point scale, where 5 meant ”a very safe person”, 4 ”a rather safe person”, 3 ”a person who is neither safe
nor unsafe”, 2 ”a rather unsafe person”, and 1 ”a very unsafe person”. In order to compare the groups on memory bias, the target faces were
divided into three categories for each person, depending on his or her rating of the degree of safety of the faces seen. Faces rated as 4 or 5
were categorized as safe, faces rated as 3 were categorized as neutral, and faces rated as 2 or 1 were categorized as unsafe. The proportion of
safe, neutral and unsafe faces that were recognized was then computed for each subject.
3
This point is worth repeating, since the face recognition paradigm has been criticized by MacLeod (1999) on the ground that “the ‘threat-
related stimuli’ employed were selected because of their special relevance to the patients’ domains of concern, and may not have been at all
threatening for the non-anxious control subjects” (p. 459). But, on the contrary, the stimuli were not selected because they had any special
relevance to the patients’ domains of concern; the photos were not even selected to be threat-related at all!
4
For example, the interpretation “being criticized” is central to the fear network that characterizes social phobia, and faces that are
interpreted as critical by social phobics are therefore likely to activate this fear structure and to be more extensively scanned than other faces
– and thereby easier to recognize. Since the fear structure that characterizes agoraphobia does not involve an equally central role for the
interpretation “being criticized”, no similar memory bias for critical faces was found among patients with this disorder. On the other hand,
when the encoding task led to an activation of interpretations in terms of people being “safe” helpers or not, a memory bias for safe faces
 If different kinds of threat-relevant encoding tasks lead to memory biases in different anxiety disorders,
then there are probably encoding tasks that lead to a memory bias for threat-related stimuli also in individuals
with acute PTSD. If crime victims with acute PTSD are given an encoding task that involves a judgment of
persons on photos in terms of their potential dangerousness, then their pathological trauma memory structure is
likely to become activated when they see persons that they interpret as potentially dangerous. This may in turn
lead to a more extensive scan of such faces. As a consequence, more information will be encoded for these faces
than for faces judged as not potentially dangerous. This should lead to a memory bias for faces that are threat-
relevant to crime victims with acute PTSD. Furthermore, since the trauma memory structure is internally
cohesive and unintegrated with pre-trauma schematic memory elements, PTSD Ss should display an inferior
memory for stimuli that are interpreted as non-dangerous.
The primary purpose of the present paper was to extend the face recognition paradigm to crime victims
with acute PTSD, and to test the hypothesis that these patients will show a memory bias for faces that are
interpreted in terms of their most basic concerns. It is a basic concern of crime victims with acute PTSD that they
are afraid of becoming perpetrated by others again. One way of activating these persons’ fears while watching
photos of unfamiliar persons is to let them rate to what extent they think that the person on the photo could
behave in a hostile way towards them. To the extent that this produces an activation of a trauma memory
structure and a heightened attention to perceived signs of hostility in other people, this might be expected to lead
to a more elaborated encoding of faces that are perceived as dangerous in crime victims with acute PTSD, and
therefore to better recognition of such faces than of faces that are not judged as dangerous. Secondly, if this
effect is specific to crime victims with acute PTSD it should not be found in a healthy control group.
Another important question was raised because of the concentration impairments that has been found to
characterize this disorder. This is hypothesized to be due to the loose integration and incompatibility of the pre-
trauma schema memory elements with the elements in the trauma memory network. It is possible that these
patients’ concentration impairments will lead them to elaborate less information on the photos, and therefore to
be less able to recognize the persons on the photos later on. If so, we should expect crime victims with acute
PTSD to show poorer memory performance than healthy controls on the face recognition task. McNally et al.
(1998) found that PTSD patients exhibited recall deficits for positive and neutral words, but not for trauma-
related words, as compared with a matched control group, and explained this in terms of their impaired
concentration; “we suspect that concentration impairments among the PTSD participants reduce processing
resources such that trauma-related material is encoded at the expense of other material” (p. 600). If this is true
also of crime victims with acute PTSD, a similar reduced level of performance should be expected on the face
recognition task.
To summarize: Two different hypotheses are tested by the present study: (1) Because of the
incompatibility of the pre-trauma memory elements with the trauma memory network the PTSD patients will
perform less well on faces that are interpreted as non-dangerous on the face recognition task, i.e., they will
recognize less non-dangerous faces than the control group. (2) The PTSD Ss will allocate their attentional
resources primarily to faces that they perceive as dangerous, so that they will recognize a larger proportion of
these faces than of faces that they do not perceive as threatening; the control group, on the other hand, will not
show this kind of memory bias.

was found among the agoraphobics. “Safety” may be assumed to be a central interpretive element in the fear structure that characterizes
agoraphobia, and faces categorized as “safe” may therefore motivate the person to a more extensive scanning and a better recognition
memory for these faces.