Current Biology, Vol. 15, 2256–2262, December 20, 2005, ª2005 Elsevier Ltd All rights reserved.DOI 10.1016/j.cub.2005.10.072
The Neural Basis of the Behavioral
Galit Yovel,1,2,* and Nancy Kanwisher1
1McGovern Institute for Brain Research
Department of Brain and Cognitive Sciences
Massachusetts Institute of Technology
Cambridge, Massachusetts 02139
Two of the most robust markers for ‘‘special’’ face pro-
the disproportionate drop in recognition of upside-
down (inverted) stimuli relative to upright faces—and
the face-selective fMRI response in the fusiform face
area (FFA). However, the relationship between these
that the behavioral FIE is closely associated with the
fMRI response in the FFA, but not in other face-
selective or object-selective regions. The FFA and the
face-selective region in the superior temporal sulcus
(f_STS), but not the occipital face-selective region
(OFA), showed a higher response to upright than in-
verted faces. However, only in the FFA was this fMRI-
FIE positively correlated across subjects with the
behavioral FIE. Second, the FFA, but not the f_STS,
showed greater neural sensitivity to differences be-
gesting a possible neural mechanism for the behav-
ioral FIE. Although a similar trend was found in the
occipital face area (OFA), it was less robust than the
FFA. Taken together, our data suggest that among
the face-selective and object-selective regions, the
FFA is a primary neural source of the behavioral FIE.
Results and Discussion
The present study tested the role of each of the three
face-selective regions (FFA, OFA, and f_STS), plus an
object-selective region called the lateral occipital com-
We predicted that cortical region(s) underlying the be-
havioral FIE should show a close association between
fMRI responses and behavioral discrimination ofupright
and inverted faces. In particular, our first prediction was
that the difference in fMRI response to upright and in-
and inverted faces (behavioral FIE). Second, a cortical
region involved in the behavioral FIE would be expected
to show greater sensitivity to differences between faces
is involved in the FIE will show a greater fMR-adaptation
effect [3, 4] when faces are upright than inverted.
Behavioral and fMRI measures of the FIE were col-
lected in two experiments. In one experiment (n = 21),
we used an event-related design in which upright- and
inverted-face trials were presented randomly within
the same scan. In a second experiment (n = 14), we
used a blocked design where upright- and inverted-
face trials were presented in separate blocks.
In both experiments, subjects performed a task dur-
ing the main fMRI scans in which they were asked
whether the two sequentially presented faces in each
wereeither bothupright orbothinverted (seeFigure 1A).
Behavioral data showed a significant drop in accuracy
for inverted compared to upright faces (i.e., a behavioral
FIE): event-related experiment: upright: 82%, inverted:
upright: 77%, inverted: 69%, t(13) = 4.29, p < .001.
Functional MRI Analyses
Each subject participated in a blocked localizer scan
(see Figure 1B) in the same scanning session as the
main experiment, in which they viewed faces, objects,
(p < 1024, uncorrected) served to functionally identify in
each subject the following regions of interest (ROIs)
across the two experiments: FFA (in 100% of subjects),
The LOC was identified in 100% of subjects with a com-
parison ofobjects >scrambled objects (p <1024, uncor-
In a few of the subjects in the event-related experi-
ment, the face-selective areas that were revealed by
the localizer did not show a higher response to faces
than nonfaces (i.e., chairs) in the experimental task
(see Stimuli section), which suggests that the localizer-
defined face-selective regions were not reliable. Thus,
only subjects that showed higher responses to faces
in the localizer-defined face-selective regions were in-
cluded in the analyses (n = 17 for FFA, n = 14 for OFA,
and n = 11 for f_STS). All the subjects in the blocked-
design task showed higher response to faces than non-
faces (houses) in the face-selective regions that were
defined by the localizer.
Main Experimental Task
Having identified the ROIs in each subject based on the
localizer data, we then calculated the magnitude of the
ditions of the main experiments. These data were used
to test whether each of the four ROIs is involved in gen-
erating the behavioral FIE. Analyses showed no differ-
ence in the pattern of response between the right and
left hemispheres for each of the ROIs. Therefore, all
analyses are based on a pooled analysis in which right
and left hemisphere voxels are combined in each ROI.
2Present address: Department of Psychology, Tel Aviv University,
Tel Aviv 69978, Israel.
1. Functional MRI Face-Inversion Effect. Assessment
ofthe inversion effectineach ROIshowed asignificantly
[t(30) = 4.67 p < .0001] and f_STS [t(21) = 3.89, p < .001],
but not in the OFA [t(27) = .98, p > .33], and an opposite
in LOC [t(34) =2.31, p <.03]. ANOVA with area and orien-
tation as repeated measures showed that the inversion
effect was significantly larger in the FFA than the OFA
[F(1,26) = 6.25, p < .02] and significantly different from
the LOC [F(1,28) = 32.31 p < .00001] but not from the
f_STS (see Supplemental Data available with this article
online for an omnibus ANOVA).
Thus, the FFA and f_STS showed a higher overall re-
sponse to upright than inverted faces, whereas the
OFA showed no significant difference and the LOC
showed an opposite effect of a higher response to in-
verted than upright faces. These analyses might seem
to implicate the f_STS and FFA (and perhaps also the
LOC) in the behavioral face-inversion effect. However,
the mere existence of an fMRI-FIE need not imply that
it is necessarily related to the behavioral FIE. We there-
fore went on to conduct a stronger test of the relation-
ship between the fMRI-FIE and the behavioral FIE.
2. A Correlation across Subjects between fMRI and
Behavioral Measures of the Face-Inversion Effect. We
Figure 1. Procedure of the Main and Localizer Tasks
(A) Experimental task. In both experiments, subjects were presented with a sequential matching face task. The upright and inverted trials were
mixed in the event-related study and in separate blocks in the blocked-design study. The stimuli presented in the figure are sample stimuli from
the event-related experiment.
(B) LocalizertTask. Subjects were presented with 16 s blocks of faces, objects, and scrambled images of objects (bodies and scenes were also
presented butare not included intheanalysis). Each block included 20 stimuli,and subjects were instructed topress akey whenever twoimages
were repeated consecutively (one-back task). We localized face-selective regions (face > object) and object-selective regions (object > scram-
bled objects). For more details about the localizer, see .
fMRI and Behavioral Face-Inversion Effect
Figure 2. The Response of Each of the Three Face-Selective Regions and the Object-Selective Region to Upright and Inverted Faces
tested whether the fMR-FIE was correlated across sub-
jects with the behavioral FIE as follows. We calculated
for each ROI in each subject. In addition, a behavioral
for upright 2 inverted): in the face-matching tasks. We
then measured the correlation across subjects between
these behavioral and neural FIE scores (see Figure 2,
right column), separately for each ROI. To increase the
power of the correlation analysis, we included data
from both the event-related and the blocked-design ex-
periments in this analysis.
was positively correlated with the behavioral FIE [r(28) =
.50, p < .005]. Interestingly, the f_STS, which showed a
higher response to upright than inverted faces in mean
response, did not show a correlation between the
f_STS-FIE and the behavioral FIE [r(18) = .06, p > .79].
This finding is consistent with the hypothesis that the
f_STS is specialized for representing dynamic aspects
of facial information such as expression and gaze [5, 6],
not identity information , which was measured with
our behavioral task. The OFA, which showed no differ-
ence between the response to upright versus inverted
faces in group means, also failed to show a correlation
between the OFA-FIE and the behavioral FIE [r(26) =
2.004, p > .98]. These findings are consistent with a re-
cent report, which suggests that the OFA is sensitive to
physical aspects of the face stimulus rather than to
face identity .
Finally, previous studies suggested that the higher re-
sponse to inverted than upright faces in house-selective
areas(i.e., areas thatshowedhigher response tohouses
than faces), which may partly overlap with the ventral
FIE [9, 10]. In contrast to this hypothesis, we found that
the higher response to inverted than upright faces in the
object-selective LOC (LOC-FIE) was not correlated with
the behavioral FIE [r(32) = .05, p > .77]. Notably, the lack
of correlation between the behavioral FIE and the OFA-
FIE or f_STS-FIE can not be due to low reliability of the
fMRI-FIE measures from these two face-selective re-
gions: the f_STS-FIE and OFA-FIE were strongly corre-
lated with each other [r(19) = .70, p < .01], which suggest
that the FIE inthese regions were highly reliable and that
their lack of correlation with the behavioral FIE reflects
lacks of association rather than low reliability. The zero
order correlations between the fMR-FIE of all face-
selective regions are reported in Table 1 (see Supple-
mental Data for correlational analyses with a normalized
FIE measure and a multiple regression analysis that as-
sesses that contribution of each face-selective region to
In summary, our analyses indicate that among the
face-selective and object-selective regions, only the
FFA-FIE is related to the behavioral FIE. Next, we used
event-related fMRI adaptation to test whether the fMR
response is more sensitive to differences between faces
when they are upright than when they are inverted.
3. Neural Discrimination of Identity Information in Up-
right and Inverted Faces. The most straightfoward ac-
count of the behavioral FIE is that neural populations
in face-processing regions discriminate individual iden-
tity of faces better when they are upright than when they
are inverted. To assess whether each ROI is more sensi-
tive to differences between faces when they are pre-
sented upright than inverted, we measured the fMR ad-
separately analyzing the fMRI response on trials in
which two different faces were presented from trials in
which the same face was presented twice. A higher re-
sponse on different than same trials in a given region
would indicate discrimination of the two different stimuli
by neural populations within that region . An interac-
tion of orientation and adaptation (different/same),
which reflects a higher response on different than
same trials for upright faces, but not for inverted faces,
was found for the FFA [F(1,16) = 7.93, p = .01] and OFA
[F(1,13) = 11.04, p < .01], but not the f_STS or LOC (see
Figure 3). These findings are indicatative of better neural
discrimination of upright than inverted faces in the FFA
and OFA. (See Supplemental Results for analyses of
and analyses based on only correct responses.)
In sum, although the OFA-FIE showed similar re-
sponse to upright and inverted faces and was not corre-
lated with the behavioral FIE, the stronger adaptation to
upright than inverted faces in the OFA suggests that it
does treat upright and inverted faces differently. Thus,
our findings with respect to the role of the OFA in the be-
havioral FIE are not conclusive. Importantly, both the
correlation and the adaptation findings clearly suggest
that the f_STS do not play a role in the behavioral FIE
in an identity discrimination task, which is consistent
(e.g., gaze, expression), rather than static (e.g., identity)
aspects of face processing . Finally, in contrast to
a higher response to inverted than upright faces (middle column). Error bars represent the standard error of the difference between upright and
than the response of the f_STS and LOC (w.4). For each ROI, we calculated the correlation across subjects between the fMRI-FIE and the be-
havioral FIE. The correlation is represented in a scatterplot for each ROI (right column) in which each subject is represented by a point, with that
subject’s fMRI-FIE score for each ROI on the x axis and the same subject’s behavioral FIE score on the y axis. Only the FFA-FIE was correlated
with the behavioral FIE.
Table 1. The Zero-Order Correlations between the fMRI Face-
Inversion Effects of the Three Face-Selective ROIs and the
n = 26
p < .05
n = 21
p < .02
n = 19
p < .001
n = 30
p > .38
n = 27
p > .25
n = 22
p > .45
Correlation was calculated by the following formula: % signal
change_upright 2 % signal change_inverted.
fMRI and Behavioral Face-Inversion Effect
prior claims [9, 10] that the higher response to inverted
than upright faces in the LOC might be related to the be-
havioral FIE [9, 10], both the correlation and the adapta-
tion data in our study suggest that the LOC plays no role
in the behavioral FIE.
Our findings indicate that among the face-selective and
object-selective regions, the FFA is a primary neural
source of the behavioral FIE. First, the FFA showed
a higher response to upright than inverted faces (FFA-
Figure 3. fMR Adaptation to Upright and In-
verted Faces in Each of the Face-Selective
Regions and the Object-Selective Region
The fMRI response of each of the face-selec-
tive (FFA, OFA, and f_STS) and object-selec-
tive (LOC) regions to pairs of different versus
same faces that were presentedupright or in-
verted. A higher response to different than
same pairs indicates neural discrimination
between faces. The FFA and OFA showed
better discrimination between upright than
between inverted faces. The f_STS and LOC
did not discriminate either upright or inverted
faces. Error bars represent the standard error
of the difference between the response to dif-
ferent and same trials. Note that the scale of
the y axis is not the same for all three regions
because the response in the OFA (w.8) to
faces is much larger than the response of
the f_STS and LOC (w.3).
FIE) that was positively correlated with the behavioral
FIE across subjects. Second, the FFA was more sensi-
tive to differences between faces when they were pre-
sented upright then when presented upside-down.
These findings are important for three reasons. First,
they resolve a long-standing mystery of the relationship
between the two most well-established markers of face
processing, the behavioral FIE and the FFA, providing
strong evidence that it is neural representations in the
tions on upright faces [8, 12] but not inverted faces. Sec-
ond, our findings provide a possible explanation of the
behavioral FIE in neural terms: the lower behavioral sen-
sitivity to differences between inverted faces compared
to upright faces apparently results at least in part from
sharper neural tuning for upright than inverted faces.
Third, our results show clear functional dissociations
between the FFA and other cortical regions engaged in
face and object perception.
Four previous reports that examined the fMRI re-
sponse to upright and inverted faces reported either
no FIE in the FFA or a weak effect, whereas our study re-
vealed a robust FIE in the FFA. Although the precise rea-
sons for these varying results are not yet clear, we spec-
ulate that the fMRI-FIE is larger and more robust for
tasks that generate a strong behavioral face inversion
effect, which was not included in most previous reports.
What Is the Nature of the Representation
of Upright Faces?
By linking the FFA to the classic behavioral hallmark of
special face processing, the behavioral FIE, the present
ture of the representations underlying face recognition.
Here again behavioral work provides important clues.
Numerous behavioral studies have indicated that what
is ‘‘special’’ about the processing of upright but not in-
verted faces is that only upright faces are processed
‘‘holistically,’’ such that face parts are processed inter-
actively rather than independently (for review, see
). Indeed, arecent fMRI study that used the compos-
ite face task , which provides a behavioral measure
of holistic processing, found a neural correlate of this ef-
fect in the FFA (C. Schiltz and B. Rossion, 2005, Vision
Sci. Soc., abstract).
However, fMRI investigations are of course severely
limited in both their spatial and temporal resolution,
leaving open many questions about the neural mecha-
nisms responsible for the differences in the FFA re-
sponse to upright and inverted faces. Do the inversion
effects reported here reflect the operation of a single
neural population in which most neurons respond
more strongly to upright than inverted faces [15, 16], or
does the FFA contain distinct neural populations, one
responsive only to upright faces and another responsive
only to inverted faces? Evidence for at least partly sep-
arate representations for upright and inverted faces
comes from a behavioral study that found opposite fig-
ural aftereffects simultaneously induced on upright and
inverted faces . Another important dimension of the
FIE unaddressed here concerns the temporal profile of
the processing of upright and inverted faces. Event-
related potential studies in humans  and single-cell
recordings in macaques  have found that latencies
of neural responses to inverted faces are delayed
compared to upright faces. It will be important for future
studies to test whether the fMRI signature of the FIE re-
ported here is associated with the differences in re-
sponse to upright and inverted faces that are found in
To sum up, the present findings have advanced our
understanding of the neural basis of face processing
by clearly linking fMRI and behavioral markers of spe-
cialized face processing (namely, the FFA and the be-
havioral FIE), by functionally dissociating the FFA from
the OFA, F_STS, and LOC, and by suggesting a possible
neural mechanism underlying
(namely, sharper neural tuning for upright than inverted
faces). Continued progress in understanding how faces
are represented inthe brainis likely toresult from further
fMRI investigations motivated by the rich behavioral lit-
erature on face processing.
Event-Related FIE Experiment
Eighteen different faces were generated by FACE 3.0 software.
There were six sets of three faces that shared the same hair but dif-
fered in internal facial information and face outline. In each trial, the
facial information, rather than hair per se, to identify the stimuli. The
selectivity to faces (faces > chairs) in the voxels that were identified
as face selective by the localizer (see Localizer Results).
The experiment consisted of five runs of the experimental task and
five runs of the localizer, which were presented interleaved with
the experimental task.
Experimental Task. Pairs of face or chair stimuli were presented
sequentially either upright or inverted in a randomized order that
was optimized for the extraction of the hemodynamic response in
an event-related fast presentation design. Each trial lasted 2 s.
The first and second stimuli were presented for 250 ms, with an in-
terstimulus interval of 500 ms. The next trial was presented 1 s after
the disappearance of the second stimulus. Five runs of the experi-
mental task were included. Each run included 24 trials of each con-
dition. In half of the trials, the stimuli were identical, and in the other
half, they were different. Each stimulus was presented equally often
intheidenticalanddifferent conditions,soanydifferenceinthe fMRI
response between conditions must be due to the relationship be-
tween the two stimuli in a pair, not the individual stimuli themselves.
Same-different responses were made with a key press. Twenty-four
blank trials were presented intermixed with the experimental trials.
Each scan lasted 240 s. Subjects made a same/different response
ioral measure of the face-inversion effect. For information about the
localizer scan, see Figure 1B and .
fMRI Data Acquisition
Scanning was done on a 3T Allegra scanner at the MGH/MIT/HMS
Athinoula A. Martinos Center for Biomedical research in Charles-
town, MA. A head coil and a Gradient Echo pulse sequence with
TR 2 s, TE 30 ms: flip angle 90º were used. Twenty-eight 4 mm thick
slices that covered the entire brain were oriented parallel to the tem-
fMRI Data Analysis
Data were analyzed separately for each subject with FS-fast (http://
surfer.nmr.mgh.harvard.edu/). Motion correction was conducted
prior to data analysis with the AFNI motion correction algorithm
 to align all the images to the first time image of the first run.
Blocked Design (Localizer). Data from the localizer were spatially
smoothed with a Gaussian filter (full width half maximum = 5 mm)
and were used to define the ROIs separately for each subject. A g
function with delta = 2.25 and tau = 1.25 was used to estimate the
fMRI and Behavioral Face-Inversion Effect
hemodynamic response (HDR) for each condition in the localizer
Event-Related Design (Experimental Task). A deconvolution anal-
ysis was used for the analysis of the event-related experimental task
to extract the HDR for each same and different pair of stimuli in the
event-related adaptation task, and no assumption was made on the
shape of the HDR. The data from the experimental runs were not
smoothed. Statistical analysis was performed on the peak of the he-
modynamic response. The time points used as the peak were deter-
mined in a hypothesis-neutral fashion based on the shape of the re-
sponse ineach area in each subject to the average of all the stimulus
conditions; the peak time points included either the 4thTR (6–8 s) or
the average of the 4thand 5thTRs, when the peak lasted more than
Blocked FIE Experiment
Theexperimentaltask included upright andinvertedfacestimuliand
upright house stimuli. Here we focus only on the face stimuli. The
face stimuli subtended 2.5º (width) 3 4º (length) of visual angle.
The face stimuli differed by either spacing among the parts or the
shape of parts and were presented in different blocks in upright or
inverted orientation (the mean response of the FFA across subjects
to the part and spacing condition for upright and inverted faces are
reported in ). Our findings showed no difference in fMRI re-
sponse and in fMRI-FIE to the spacing and part face stimuli. There-
fore, for the purpose of the investigation of the FIE, we averaged the
response across the two face sets.
The experiment included five runs of a localizer (see Figure 1B) and
six runs of the experimental task. Upright and inverted face stimuli
were presented in separate blocks. Each block included 10 trials
in which a pair of faces that were either identical (on 50% of trials)
or different were presented sequentially. Subjects performed
asame-different discriminationtaskoneach pairofstimuli bypress-
ing one key for a ‘‘same’’ response and another key for a ‘‘different’’
response. The order of the upright and inverted blocks was counter-
balanced across scans.
fMRI Data Acquisition and Analysis
See methods for event-related experiment.
Supplemental Data include one table and Supplemental Results and
can be found with this article online at http://www.current-biology.
We would like to thank Chris Baker, Brad Duchaine, Johannes
Haushofer, Margaret Livingston, Elinor McKone, Hans Op de Beeck,
Rachel Robbins, Rebecca Schwarzlose, and Doris Tsao for their
comments on the manuscript. We also thank Ming-fai Fong for
help with data analyses. This research was supported by NIH grants
66696 and EY13455 to N.K., by the National Center for Research Re-
sources (P41-RR14075, R01 RR16594-01A1, and the NCRR BIRN
Morphometric Project BIRN002), and by the Mental Illness and Neu-
roscience Discovery (MIND) Institute.
Received: August 6, 2005
Revised: October 28, 2005
Accepted: October 31, 2005
Published: December 19, 2005
1. Malach, R., Reppas, J.B., Benson, R.R., Kwong, K.K., Jiang, H.,
Kennedy, W.A., Ledden, P.J., Brady, T.J., Rosen, B.R., and Too-
tell, R.B. (1995). Object-related activity revealed by functional
magnetic resonance imaging in human occipital cortex. Proc.
Natl. Acad. Sci. USA 92, 8135–8139.
2. Kanwisher, N., Chun, M.M., McDermott, J., and Ledden, P.J.
(1996). Functional imaging of human visual recognition. Cogn.
Brain Res. 5, 55–67.
3. Grill-Spector, K., and Malach, R. (2001). fMR-adaptation: a tool
forstudyingthefunctional properties ofhumancorticalneurons.
Acta Psychol. (Amst.) 107, 293–321.
4. Kourtzi, Z., and Kanwisher, N. (2001). Representation of per-
ceivedobjectshapebythehumanlateraloccipital complex. Sci-
ence 293, 1506–1509.
5. Puce, A., Allison, T., Bentin, S., Gore, J.C., and McCarthy, G.
(1998). Temporal cortex activation in humans viewing eye and
mouth movements. J. Neurosci. 18, 2188–2199.
6. Hooker, C.I.,Paller, K.A., Gitelman,D.R., Parrish, T.B.,Mesulam,
M.M., and Reber, P.J. (2003). Brain networks for analyzing eye
gaze. Brain Res. Cogn. Brain Res. 17, 406–418.
7. Hoffman, E.A., and Haxby, J.V. (2000). Distinct representations
of eye gaze and identity in the distributed human neural system
for face perception. Nat. Neurosci. 3, 80–84.
8. Rotshtein, P., Henson, R.N., Treves, A., Driver, J., and Dolan,
R.J. (2005). Morphing Marilyn into Maggie dissociates physical
and identity face representations in the brain. Nat. Neurosci. 8,
9. Haxby, J.V., Ungerleider, L.G., Clark, V.P., Schouten, J.L., Hoff-
man, E.A., and Martin, A. (1999). The effect of face inversion on
activity in human neural systems for face and object perception.
Neuron 22, 189–199.
10. Rossion, B., and Gauthier, I.I. (2002). How does the brain pro-
cess upright and inverted faces? Behav. Cogn. Sci. Rev. 1, 63–
11. Haxby, J.V., Hoffman, E.A., and Gobbini, M.I. (2000). The distrib-
uted human neural system for face perception. Trends Cogn.
Sci. 4, 223–233.
12. Grill-Spector, K., Knouf, N., and Kanwisher, N. (2004). The fusi-
form face area subserves face perception, not generic within-
category identification. Nat. Neurosci. 7, 555–562.
13. Tanaka, J.W., and Farah, M. (2003). The holistic representation
of faces. In Perception of Faces, Objects and Scenes: Analytic
and Holistic Processes, M.A. Peterson and G. Rhodes, eds.
(New York: Oxford University Press), pp. 53–74.
14. Young, A.W., Hellawell, D., and Hay, D.C. (1987). Configurational
information in face perception. Perception 16, 747–759.
15. Perrett, D.I.,Oram,M.W., and Ashbridge, E.(1998).Evidence ac-
cumulation in cell populations responsive to faces: an account
of generalisation of recognition without mental transformations.
Cognition 67, 111–145.
16. Tanaka, K., Saito, H., Fukada, Y., and Moriya, M. (1991). Coding
visual images of objects in the inferotemporal cortex of the ma-
caque monkey. J. Neurophysiol. 66, 170–189.
17. Rhodes, G., Jeffery, L., Watson, T.L., Jaquet, E., Winkler, C., and
Clifford, C.W. (2004). Orientation-contingent face aftereffects
and implications for face-coding mechanisms. Curr. Biol. 14,
18. Bentin, S., Allison, T., Puce, A., Perez, E., and McCarthy, G.
(1996). Electrophysiological studies of face perception in hu-
mans. J. Cogn. Neurosci. 8, 551–565.
19. Perrett, D.I., Smith, P.A., Potter, D.D., Mistlin, A.J., Head, A.S.,
Milner, A.D., and Jeeves, M.A. (1985).Visual cells inthe temporal
cortex sensitive to face view and gaze direction. Proc. R. Soc.
Lond. B. Biol. Sci. 223, 293–317.
istration for functional MRI. Magn. Reson. Med. 42, 1014–1018.
21. Yovel, G., and Kanwisher, N. (2004). Face perception: domain
specific not process specific. Neuron 44, 889–898.