Abstract
Two identical top parts of a face appear distinct if their respective bottom parts differ. This notorious ‘composite face effect’ (Young et al., 1987) is taken as strong evidence for the holistic nature of face perception, i.e. the fact that faces are processed as undecomposed wholes. Here we tested the hypothesis that facial identity is coded holistically by face-sensitive neurons in the middle fusiform gyrus (MFG), where the largest sensitivity to face stimuli has been disclosed in neuroimaging. We used an fMR-adaptation design to compare activity in the MFG while 8 healthy normal subjects viewed blocks of face stimuli in which the top and bottom parts of the faces were either aligned or misaligned. Whereas the top of the face presented in a block was always the same, the bottom parts were identical in the ‘same’ condition, but differed from trial to trial in the ‘different’ condition. The design thus consisted of 4 conditions: mis/alignment X same/different bottom parts. The subject's task was to concentrate on the top part and detect rare colour changes occurring at this location. In the face-sensitive area in the right MFG we observed a significant recovery from adaptation to facial identity in the ‘different’ compared to the ‘identical’ condition, but only when the top and the bottom parts were aligned. This neuronal response pattern is consistent with the illusion of viewing different faces, which occurs exclusively in the ‘different’ blocks consisting of aligned face parts. The neuronal composite effect was also significant, albeit slightly weaker, in the face sensitive areas in the left MFG as well as bilaterally in the inferior/lateral occipital gyrus. Based on these results we hypothesize that the coarse processing underlying holistic face processing takes place first in the MFG and subsequently serves as a header to the processing of detailed features, which relies on the integration of information from the MFG and the inferior/lateral occipital gyrus.