BEHAVIOURAL

BRAIN
RESEARCH
ELSEVIER BehaviouralBrain Research 76 (1996) 181-189

Is face recognition 'special'? Evidence from neuropsychology

M a r t h a J. F a r a h *

University of Pennsylvania, Philadelphia, PA, USA

Received 17 December 1994; accepted 14 March 1995

Abstract

Is face recognition 'special,' in the sense of relying on functionally and anatomically distinct mechanisms from those required
for other kinds of pattern recognition? A number of different neuropsychological dissociations involving recognition and learning
of faces and nonface object~,~are reviewed. In addition, studies of the nature of shape representation in normal face and object
recognition are reviewed. The evidence from brain-damaged and normal subjects suggests that face recognition is, indeed, 'special,'
and provides some clues to ~:hefunctional differences between face and object recognition.

Keywords: Face recognition;Neuropsychologicaldissociation;Object recognition

I. Introduction tion, such as the distinction between object and location

representation or the use of viewer-centered versus
The recognition of faces and many other types of object-centered representations, yet confine their set of
three-dimensional objects seem to pose essentially the objects to faces (e.g., [11,15]).
same problem. Depending upon the angle from which In contrast, other lines of research have focused on
we view them, or the positions of their moveable parts, the ways in which face recognition may be different from
they may cast radically different images on our retinae, common object recognition. In addition to the neuro-
as demonstrated in Fig. 1. The primary task our visual psychological support to be discussed in this chapter,
systems confront in recogaizing these stimuli is to create evidence from normal subjects suggests that face recogni-
a representation that discriminates among them yet is tion is, at least in some ways, different from other types
invariant over at least a range of viewing conditions of object recognition. For example, infants are born with
(e.g., [ 12]). When thought about in this way, the recogni- a preference to gaze at faces rather than other objects.
tion of common objects and of faces seem to pose largely At just 30 min of age, they will track a moving face
the same problem. Indeed, Marr [12] discussed them farther than other moving patterns of comparable con-
together and suggested an explanation of the effects of trast, complexity, and so on (see [14] for a review of
orientation on terms of his theory of object recognition this and other studies of infant face perception). The
(p. 310). The assumption that face and object recognition effects of orientation on face recognition, already men-
require essentially the sa~aae mechanisms can be found tioned, provides another indication that face recognition
in many different disciplines in which vision is studied. is special. Whereas most objects are a bit harder to
Within neuropsychology,, for example, Damasio and recognize upside down that rightside up, inversion makes
colleagues have asserted that face recognition is an faces dramatically harder for normal adult subjects to
instance of the more general problem of recognizing a recognize (see [ 18] for a review).
particular exemplar of an object category (e.g., [2]), be
These findings from normal subjects indicate differ-
it one person's face amo~ag all faces, or one particular
ences between face recognition and the recognition of
chair among all chairs. Within neurophysiology,
other objects: Face recognition has earlier developmental
researchers may address general issues in object recogni-
precursors and is more orientation-sensitive than the
recognition of other types of object. However, they do
* Department of Psychology, 3815 Walnut St., Philadelphia, PA not necessarily imply that face recognition is accom-
19104-6196, USA. plished using a different system from object recognition.

0166-4328/96/$15.00 © 1996ElsevierScienceB.V. All rights reserved

SSDI 0166-4328(95)00198-0
182 Martha J. Farah/Behavioural Brain Research 76 (1996) 181-189

Fig. 1. Examplesof how images of faces and other objects vary accordingto viewpoint.

Faces could be the first type of shape that a general- For systems to be considered different, they should be
purpose system represents. Similarly, faces could require functionally independent, such that either can operate
a special orientation-sensitive type of shape representa- without the other. They should also be physically distinct
tion that is derived within a physically unitary and (not necessarily redundant with the first criterion, as two
functionally indivisible system. programs running on the same computer could be
The hypothesis of interest in the present article is functionally independent). Finally, they should not be
whether or not face and nonfaee object recognition are mere physical and functional duplicates of one another,
accomplished by different systems. I will therefore begin but should process information in different ways. In the
by stating explicitly what I mean by 'different systems.' present article, I will review neuropsychological evidence
 Martha J. Farah/BehaviouralBrain Research 76 (1996) 181-189 183

relevant to the hypothesis that face recognition depends prosopagnosics have some degree of difficulty recogniz-
upon a different system from the recognition of ing objects other than faces, in some cases the deficit
common objects. appears strikingly selective for faces.

2.1. Is prosopagnosia really selective for faces

2. Visual agnosia and prosopagnosia
The most straightforward interpretation of prosopag-
The most relevant neu:rosychological impairment for nosia is that the highest levels of visual representation
present purposes is visual agnosia. The term agnosia are subdivided into specialized systems, and prosopag-
refers to an impairment of object recognition that is not nosics have lost the system that is necessary for recogniz-
attributable to a loss of general intellectual ability or to ing faces and not necessary, or less necessary, for
an impairment in elemenlary perceptual processes, such recognizing other types of object. However, it is possible
as brightness, acuity, dep~Ih, or color vision (see [4] for that faces and other types of object are recognized
a detailed overview). Thus by definition, agnosics retain using a single recognition system, and that faces are
full knowledge of the nc,nvisual aspects of the object, simply the most difficult type of object to recognize.
enabling them to recognize it by touching it, hearing its Prosopagnosia could then be explained as a mild form
characteristic sounds, or hearing a verbal definition of of agnosia, in which the impairment is detectable only
it. They can also perceive at least some of its visual on the most taxing form of recognition task. This
properties. account has the appeal of parsimony, in that it requires
In associative agnosia ]perception can be remarkably only one, single type of visual recognition system, and
preserved, to the extent tl~at the person may be able to perhaps for this reason has gained considerable
draw a good copy of a drawing or object that he cannot popularity.
recognize. Indeed, the term was coined, in the 19th To determine whether prosopagnosia is truly selective
century, because it seemed that perception could not be for faces, and hence whether the human brain has
at fault in these cases, and that the problem must specialized mechanisms for recognizing faces, we must
therefore lie in associating perception with knowledge therefore assess the prosopagnosic performance on faces
of the objects. Our understanding of vision has now and nonface objects relative to the difficulty of these
progressed to the point where we can identify different stimuli for normal subjects. One technical difficulty
levels of visual representation, from those early and encountered in such a project is that normal subjects
intermediate representations that make explicit the edges will invariably perform nearly perfectly on both face and
and surfaces in an image, to higher level representations nonface recognition tasks. The resultant ceiling effect
that make explicit the more stable shape properties of will mask any differences in difficulty that might exist
the distal object and thu:~ solve the 'problem' of visual between tasks, making it pointless to test normal subjects
object recognition mentioned earlier. Associative visual in the kinds of recognition tasks that have traditionally
agnosia is probably best viewed as an impairment in the been administered to patients. With this problem in
highest levels of visual representation, rather than as an mind, researchers have devised visual recognition tasks
inability to associate normal visual representations with that test learning of face and nonface objects. By having
other types of knowledge; [4]. According to this view, subjects learn to recognize specific new exemplars of
the ability of associative agnosics to draw the object faces and other types of object, it is possible to titrate
results from their use of lower level visual representa- normal subjects' level of recognition performance so that
tions, whereas recognition requires higher level represen- it falls between ceiling and floor.
tations. The observation that associative visual agnosics The first researchers to address this issue directly were
tend to copy drawings slowly and slavishly, and mistake McNeill and Warrington [13]. They studied case W.J.,
objects for visually similar objects, is suggestive of an a middle-aged professional man who became prosopag-
impairment in visual perception (see [4]). nosic following a series of strokes. After becoming proso-
Associative visual agnosia does not always seem to pagnosic, W.J. made a career change and went into
affect the recognition of all types of stimuli equally. The sheep farming. He eventually came to recognize many
selectivity observed in ~;ome cases of agnosia lends of his sheep, although he remained unable to recognize
support to the view that there are specialized systems most humans. The authors noted the potential implica-
for recognizing particular types of stimuli. The best tions of such a dissociation for the question of whether
known example of this is prosopagnosia, the inability to human face recognition is 'special,' and designed an
recognize faces after brain damage. Prosopagnosics ingenious experiment exploiting W.J.'s new-found career.
cannot recognize familia~r people by their faces alone, They assembled three groups of photographs, human
and must rely on other cues for recognition such as a faces, sheep faces of the same breed kept by W.J., and
person's voice, or distinctive clothing or hairstyles. The sheep faces of a different breed, and attempted to teach
disorder can be so severe that even close friends and subjects names for each face. Normal subjects performed
family members will not be recognized. Although many at intermediate levels between ceiling and floor in all
184 Martha J. Farah/Behavioural Brain Research 76 (1996) 181-189

conditions. They performed better with the human faces correct. As before, L.H. was disproportionately impaired
than with sheep faces, even those who, like W.J., worked at face recognition relative to nonface recognition, when
with sheep. In contrast, W.J. performed poorly with the his performance is considered relative to normal subjects.
human faces, and performed normally with the sheep Specifically, he showed significantly less face superiority
faces. These data suggest that W.J.'s recognition impair- in this task than normal subjects, achieving 64% faces
ment does not affect the recognition of all groups of correct and 63% eye glass frames correct. Like the first
visually similar patterns, but is selective for human faces. experiment, this one also suggests that L.H.'s impairment
Karen Levinson, Karen Klein and I took a similar in face recognition cannot be attributed to a general
approach, but used common objects rather than faces problem with object recognition. The present results
of another species to compare with human face recogni- also suggest that the problem does not lie with the
tion [6]. Our subject was L.H., a well-educated profes- recognition of specific exemplars from any visually
sional man who has been prosopagnosic since an homogeneous category, but is specific to faces.
automobile accident in college. L.H. is profoundly proso- A final experiment was undertaken to address the
pagnosic, unable to recognize reliably his wife, children, specificity of L.H.'s face recognition impairment. In
or even himself in a group photograph. Yet he is highly essence, the design of the previous experiments amounts
intelligent, and seems to have little or no difficulty to a comparison between a prosopagnosic's performance
recognizing other types of visual patterns such as printed with faces and his performance with stimuli that are like
words or objects. Although he has a degree of recognition faces (in their recognition difficulty, in their membership
impairment with drawings of objects, this is still less in a visually homogeneous category) without being faces,
severe than his impairment with faces. or rather without being processed by the hypothesized
We employed a recognition memory paradigm, in face-specific recognition mechanism. Stating the experi-
which L.H. and control subjects first studied a set of mental design in this way suggests the ideal nonface
photographs of faces and nonface objects, such as forks, comparison stimulus: upside-down faces. As mentioned
chairs, and eyeglasses. The photographs in Fig. 1 were earlier, inverting a face makes it much harder for normal
drawn from this experiment. Subjects were then given a subjects to recognize. On the basis of the face inversion
larger set of photographs, and asked to make 'old'/'new' effect, it is generally assumed that if a specialized face
judgements on them. This larger set was designed so recognition mechanism exists, it is specialized for the
that for each face and nonface object in the 'old' set processing of upright faces. Inverted faces therefore
there was a highly similar item in the 'new' set. For constitute ideal comparison stimuli: They are equivalent
example, there were two upholstered swiveling desk to upright faces in virtually all physical stimulus parame-
chairs with arms, one in the 'old' set and one in the ters, including complexity and inter-item similarity, but
'new.' Verbal descriptions would therefore be of minimal do not engage (or engage to a lesser extent) the hypothe-
help in performing this memory task. Whereas normal sized face-specific processing mechanisms.
subjects performed equally well with the faces and Kevin Wilson, Maxwell Drain, James Tanaka and I
nonface objects, achieving on average 85% correct, L.H. [8] reasoned that if L.H.'s underlying impairment was
showed a significant performance disparity, achieving not face-specific, then he would show a normal face
only 62% correct for faces and 92% correct for objects. inversion effect. In other words, he would perform
In a second experiment, we compared recognition of normally with upright faces relative to his performance
exemplars of the category 'face' and an equivalent on inverted faces. In contrast, if he had suffered damage
number of highly similar exemplars all drawn from a to neural tissue implementing a specialized face recogni-
single nonface category, namely eyeglass frames. tion system, he would show an absent or attenuated face
Examples of faces and eyeglass frames are shown in inversion effect. That is, he would be disproportionately
Fig. 2. The photographs were divided evenly into sets of impaired with upright faces, relative to his performance
'old' items, which would appear in the study and test on the comparison stimuli, inverted faces.
phases of the experiment, and sets of 'new' items, which L.H. and normal subjects were tested in a sequential
would appear only at test. Similar-looking eye glass matching task, in which an unfamiliar face was presented,
frames were separated into 'old' and 'new' sets to make followed by a brief interstimulus interval, followed by a
the task more challenging (e.g., there were both 'old' and second face, to which the subject responded 'same' or
'new' horn rims, and 'old' and 'new' aviator style frames). 'different'. The first and second faces of a trial were
In this experiment normal subjects found face recogni- always in the same orientation, and upright and inverted
tion considerably easier than eye glass frame recognition. trials were randomly intermixed. As expected, normal
Normal undergraduates achieved on average 87% faces subjects performed better with the upright than with the
correct and 67% eye glass frames correct. A second inverted faces, replicating the usual face inversion effect:
group of normal subjects, matched in age and education 94% and 82% correct, respectively.
level with L.H., showed the same disparity, achieving on In contrast, L.H. was significantly more accurate with
average 85% faces correct and 69% eye glass frames inverted faces! He achieved 58% correct for upright and
 Martha J. Farah/ Behavioural Brain Research 76 (1996) 181-189 185

Fig. 2. Samplestimulifroman experimenton faceand eyeglassframerecognition.

72% correct for inverted faces. This outcome was not A second conclusion that follows from these results
among the alternatives we had considered. We had concerns the 'control structure' of visual processing.
assumed that if he had an impaired face processor, it L.H.'s specialized face perception system was apparently
would simply not be used in this task, leading to the contributing to his performance even though it was
prediction of an absent or attenuated face inversion impaired and clearly maladaptive. This suggests that the
effect. Instead, it appears ]ae has an impaired face-specific specialized face system is operates mandatorily, reminis-
processor, which is engaged by the upright but not the cent of Fodor's [10] characterization of special-purpose
inverted faces, and used despite being disadvantageous. perceptual 'modules' as engaged mandatorily by their
This result was confirmed in additional studies, which inputs. The idea that the face system cannot be prevented
invariably showed either statistically significant or non- from processing faces, even when damaged, may also
significant trends in the same direction. explain why W.J. was able to learn to recognize indivi-
The 'inverted inversion effect' found in this prosopag- dual sheep after his strokes but could not learn to
nosic subject has strong implications for the selectivity recognize human faces.
of prosopagnosia. Inverte,d faces are the perfect control The general conclusion of these studies of W.J. and
stimulus for equating face:s and nonface objects for such L.H. is that prosopagnosia represents the selective loss
factors as complexity and interitem similarity. L.H.'s of visual mechanisms necessary for face recognition, and
disproportionate impairment on upright relative to not necessary (or less necessary) for other types of object
inverted faces therefore implies that an impairment of recognition. In terms of the organization of the normal
face-specific processing mechanisms underlies his visual system, these studies suggest that faces are recog-
prosopagnosia. nized differently than other objects.
186 Martha J. Farah/Behavioural Brain Research 76 (1996) 181-189

3. Selective impairment of new face learning with faces than with other objects? Second, it distingu-
ishes two possible relationships that might hold between
Individuals such as W.J. and L.H. are impaired at the specialized face system and the nonface object system,
both new face learning and recognition of previously shown in Fig. 3. The two systems could be arranged in
familiar faces, as would be expected if the substrates of series, such that all stimuli are first processed by one
face representation were damaged. Lynette Tippett, general system, and then faces receive further processing
Laurie Miller and I recently encountered someone with by the system that is specialized for faces. Alternatively,
an even more selective impairment: Case C.T. is impaired the two systems could be arranged in parallel, such that
at learning new faces, but is relatively preserved in his a stimulus is recognized either by one or the other,
recognition of previously familiar faces and in his learn- independently of one another.
ing of other nonface visual objects [ 17]. This pattern of Given the intuition that face recognition requires
performance is consistent with a disconnection between processing that is somehow more elaborate or demand-
intact face representations and an intact medial temporal ing than object recognition, which presumably motivated
memory system. As such, it provides additional evidence the alternative accounts described in the last section,
that the neural substrates of face representation are one might expect the first type of arrangement to hold.
distinct from the representation of other objects, as they According to this view there is specialized face system,
can be selectively disconnected from the substrates of but it is not functionally independent from the object
new learning. system; it requires input from the object system and
C.T.'s face perception was normal as measured by the performs further processing on that input. This arrange-
Benton and Van Allen face matching task [ 1 ]. He also ment contrasts with the first one, according to which
performed normally on the face inversion task used with earlier visual processes deliver their products to two
L.H., in terms of overall level of performance and the parallel, independent systems, one of which is required
presence of an inversion effect. His learning of verbal to recognize faces and the other of which is required to
material and even visual material other than faces is recognize other objects.
also normal. However, when given the face and eyeglass If there are, indeed, associative visual agnosics with
learning task, he performed similarly to L.H., achieving relatively spared face recognition, then the systems subs-
58% correct for faces and 63% correct for eyeglasses. erving object and face recognition must be arranged in
Additional evidence of his inability to learn faces comes parallel. In on-going work, Marlene Behrmann and I
from his identification of famous faces. For people who have confirmed experimentally the clinical observation
were famous prior to C.T.'s head injury, he performed that faces can be disproportionately spared. We used the
within the range of 8 age-matched control subjects on a same faces and eyeglasses experiment that had earlier
forced choice 'famous/not famous' task, whereas for been administered to L.H.
more recently famous individuals he performed at The subject in this experiment was C.K., a young man
chance. One celebrity allows for an especially interesting who suffered a closed head injury resulting in severe
comparison between premorbid and current face visual agnosia, with seemingly spared face recognition.
recognition: In the case of Michael Jackson, the singer's C.K.'s pattern of performance differs from normality in
extensive plastic surgery following C.T.'s injury provides the opposite direction from L.H.'s: He obtains 98%
us with a 'within-celebrity' comparison of face recogni- correct for the faces and only 48% correct for the
tion. Despite the greater popularity and media exposure eyeglasses. This result, taken together with opposite
of Michael Jackson in recent years, C.T. recognized a
picture taken in the 1970s and failed to recognize an up-
to-date photograph.
FACES

4. Preserved face recognition in visual agnosia OBJECTS FA CES

Some associative visual agnosics appear to have more

difficulty with object recognition than with face recogni-
tion, presenting us with the mirror image of the proso-
pagnosic's impaired and spared abilities [5,9]. This
o BJ ECTS
\/
pattern of impairment is interesting for two reasons. EAPt_Y
First, it offers further disconfirmation of the hypothesis VISION
EARLY
that prosopagnosia is just a mild disorder of a general- VISION
purpose object recognition system, with faces simply
being harder to recognize than other objects. If this were Fig. 3. Two possible ways that face recognition could be 'special'
true, how could it be possible for a person to do better relative to nonfaceobjectrecognition.
 Martha J. Farah/Behavioural Brain Research 76 (1996) 181-189 187

dissociation, implies that the systems specialized for face subjects to identify a set of whole faces, along with a set
and object recognition are functionally independent. Put of whole nonface objects, and then assessed their ability
more precisely, there is one system that is more impor- to recognize both the whole patterns and their parts.
tant for face recognition than for nonface object recogni- Examples of study and test stimuli are shown in Fig. 4.
tion, and another system (or systems) that is more Relative to the recognition of houses, face recognition
important for nonface object recognition than for face showed a greater disadvantage for parts relative to
recognition, and they are arranged in parallel as shown whole: Subjects achieved on average 81% and 79%
on the right side of Fig. 3. correct for parts of houses and whole houses, respec-
tively, and 65% and 77% for parts of faces and whole
faces, respectively. This is what should be expected if the
5. Functional differences between face and object representations underlying face recognition do not
recognition explicitly represent parts, or do so to a lesser degree
than nonface objects. Similar results were obtained with
Given the evidence just reviewed of at least two inverted faces and scrambled faces as the nonface com-
specialized subsystems underlying visual recognition, let parison stimuli. Further evidence that face recognition
us now turn to the question of what these specialized is distinctive by virtue of its reliance on holistic represen-
systems might be specialized for, in terms of the way in tation comes from studies of the inversion effect: We
which they represent shape information. In order to have found that the face inversion effect can be elimi-
address this question, it would be helpful to first intro- nated when faces are learned in a way that encourages
duce the concept of a structural description. Many cur- part-wise encoding, and that a factor determining how
rent theories of object recognition posit structural much the recognition of nonface patterns suffers from
descriptions, which are representations of object shape inversion is whether they were initially learned holisti-
in terms of parts, which zLrethemselves explicitly repre- cally or in terms of parts [7].
sented as shapes in their own right. The more extensive In the final experiment to be described, we bring the
the part decomposition, the more parts there will be in research back to prosopagnosia and the neural bases of
an object's representation, but the simpler those parts face recognition. The neuropsychological results
will be. The less the part decomposition, the fewer the described earlier imply that there is some neurologically
parts there will be in an object's representation, but the distinct subsystem that is more important for face
more complex those part~; will be. recognition than for other kinds of object recognition.
The conjecture being put forth here is that face The results of the experiments just described imply that
recognition is holistic, in the sense that it involves normal subjects perceive faces more holistically than
virtually no part decomposition, and hence requires the they perceive other kinds of objects. Taken together,
ability to represent complex parts. At first glance this these findings suggest that the face recognition system
hypothesis may seem already disconfirmed by the experi- damaged in prosopagnosia is a system of relatively
ments of Davidoff and Donnelly [3]. They compared holistic representation. The purpose of the final experi-
the availability of part representations during face and ment is to test this hypothesis directly.
chair perception in a sequential matching task, and Maxwell Drain, Jim Tanaka and I compared the
found no difference between faces and chairs. However, relative advantage of whole faces over face parts for
this result speaks against only the stronger hypothesis normal subjects and for the prosopagnosic L.H. Our
that subjects cannot voluntarily attend to facial parts as initial plan was to administer the same task that Tanaka
well as they can to chair parts, when stimuli are physi- and I used with the normal subjects to L.H., but despite
cally present. In collaboration with James Tanaka and intensive effort, L.H. could not learn to recognize a set
others, I have carried out several tests of the hypothesis of faces. We therefore switched to a short-term memory
that normal or 'default' fa,ce recognition differs from the paradigm, in which a face was presented for study,
recognition of other stimuli in its greater reliance of followed by a blank interval, followed by a second
holistic representation and correspondingly lesser reli- presentation of a face. The subject's task was to say
ance on part-based repre,,~entation. whether the first and second faces were the same or
In one set of studies, we reasoned as follows [16]: To different. There were two different conditions for the
the extent that some portion of a pattern is explicitly presentation of the first face: Either 'exploded' into four
represented as a part for purposes of recognition, then separate frames containing the head, eyes, nose and
when that portion is presented later in isolation, subjects mouth (in their proper relative spatial position within
should be able to identi~, it as a portion of a familiar each frame), or intact. The second face was always
pattern. In contrast, if a portion of a pattern does not presented in the normal format, so that the two condi-
correspond to the way the subject's visual system parses tions can be called 'parts-to-whole' and 'whole-to-whole'.
the whole pattern, then that portion presented in isola- Normal subjects performed better in the 'whole-to-
tion is less likely to be recognized. Tanaka and I taught whole' than in the 'part-to-whole' condition, on average
188 Martha J. Farah/Behavioural Brain Research 76 (1996) 181 189

•.~.~ ~(::
.'.~. ~:",

~;.. "~,,~-~,- .:.~

".. ~. " o. "~ ~ , ~.% •
~ .~.'r~.~ ~, ~"
~ ~. ,.

Fig. 4. Sample test stimuli from an experiment on face and house recognition.

93% and 74% correct, respectively, thus providing of face recognition in two prosopagnosic subjects,
further evidence that their perception of a whole face McNeill and Warrington's case W.J., and the subject of
is not equivalent to the perception of its parts. L.H. our studies, L.H., implies that we are endowed with a
showed abnormally little difference between the two specialized system for recognizing faces. This system is
conditions, 74% and 73% correct, respectively. This not necessary for (or is less important for) recognizing
finding is consistent with the hypothesis that he has lost objects other than human faces, even when such objects
the ability to see faces as wholes. form a large and visually homogeneous category such
as sheep faces or eyeglass frames. The selective deficit
5.1. Conclusions for new face learning observed in C.T. also supports the
existence of functional specialization for face representa-
Is face recognition 'special'? The evidence reviewed in tion. Furthermore, the system is anatomically distinct,
this article suggests that it is. The selective impairment in that it can be selectively damaged and selectively
 Martha J. Farah/Behavioural Brain Research 76 (1996) 181-189 189

disconnected from medial temporal areas by stroke [2] Damasio, A.R., Damasio, H. and Tranel, D., Impairments of
visual recognition as clues to the processes of memory. In G.
or head injury. The opposite pattern of impairment
Adelman, E. Gall and M. Cowan (Ed.), Signal and Sense: Global
observed in the agnosic :~ubject C.K. suggests that the Order in Perceptual Maps Wiley-Liss, New York, 1990.
specialized face recognition system does not merely [3] Davidoff, J. and Donnelly, N., Object superiority: A comparison
elaborate the processing of the object system, but rather of complete and part probes. Acta Psychol., 73 (1990) 225-243.
processes stimuli in parallel with it, and is at least [4] Farah, M.J., Visual Agnosia: Disorders of Object Recognition and
partially functionally independent of the other system. What They Tell Us About Normal Vision. Cambridge, MA: MIT
In addition to being physically distinct and function- Press/Bradford Books,1990.
[5] Farah, M.J., Patterns of co-occurrence among the associative
ally independent, the twc. systems also appear to differ
agnosias: Implications for visual object representation. Cogn.
in the way they represent shape. Research with normal Neuropsychol., 8 (1991) 1-19.
subjects suggests that f~tces are recognized as single [6] Farah, M.J., Klein, K.L. and Levinson, K.L., Face perception and
complex wholes, without decomposition into separately within-category discrimination in prosopagnosia. Neuropsycho-
represented parts. A study with L.H. showed that his logia, 33 (1995) 661-674.
short-term memory for faces does not benefit from the [7] Farah, M.J., Tanaka, J.R. and Drain, H.M., What causes the face
opportunity to perceive the face as a whole, consistent inversion effect? J. Exp. Psychol.: Hum. Percept. Perform, in press.
[8] Farah, M.J., Wilson, K.D., Drain, H.M. and Tanaka, J.R., The
with the idea that he has lost holistic perception of faces. inverted inversion effect in prosopagnosia: Evidence for manda-
Referring back to the issne raised at the outset, we can tory face-specific processing mechanisms. Vis. Res., in press.
now offer a tentative answer: Face recognition and object [9] Feinberg, T.E., Schindler, R.J., Ochoa, E., Kwan, P.C. and Farah,
recognition appear to depend on different systems, which M.J., Associative visual agnosia and alexia without prosopag-
are anatomically separate, functionally independent, and nosia. Cortex, in press.
differ according to the ,degree of part decomposition [10] Fodor, J.A., The Modularity of Mind. Cambridge, MA: The MIT
Press, 1983.
used in representing shape.
[11] Haxby, J.V., Grady, C.L., Horowitz, B., Ungerleider, L.G.,
Mishkin, M., Carson, R.E., Herscovitch, P., Schapiro, M.B. and
Rapoport, S.I., Dissociation of object and spatial visual process-
Acknowledgment ing pathways in haman extrastriate cortex. 35 (1995) 2089-2093.
[12] Marr, D., Vision. San Francisco: Freeman, 1982.
This research was supported by ONR grant N00014- [13] McNeil, J.E. and Warrington, E.K., Prosopagnosia: A face-
93-10621, N I M H grant R01 MH48274, NINDS grant specific disorder. Q. J. Exp. Psychol., 46A (1993).
[14] Morton, J. and Johnson, M.H., CONSPEC and CONLERN: A
R01 NS34030, Alzheimer's Association/Hearst
two-process theory of infant face recognition. Psychol. Rev., 98
Corporation Research Grant PRG-93-153, an NSF STC (1991) 164-181.
grant to the Institute for Research in Cognitive Science [15] Perrett, D.I., Oram, M.W., Hietanen, J.K. and Benson, P.J., Issues
at the University of Pennsylvania, and the University of of representation in object vision. In M.J. Farah and G. Ratcliff
Pennsylvania Research Foundation. I thank Vincent (Ed.), The Neuropsychology of High-Level Vision Hillsdale: Law-
Walsh for helpful comments and advice. rence Erlbaum Associates, 1994.
[16] Tanaka, J.W. and Farah, M.J., Parts and wholes in face recogni-
tion. Q. J. Exp. Psychol. (1993)..
[ 17] Tippett, L.J., Miller, L. and Farah, M.J., Prosopamnesia: A selec-
References tive new learning impairment for faces. (in preparation).
[ 18] Valentine, T., Upside-down faces: A review of the effects of inver-
[1] Benton, A.L., Hamsher, K, Varney, N.R. and Spreen, O., Contri- sion upon face recognition. Br. J. Psychol., 79 (1988) 471-491.
butions to Neuropsychological Assessment. New York: Oxford
University Press, 1983.