We localized regions in the fusiform gyrus and superior temporal sulcus that were more active when subjects viewed photographs of real faces than when they viewed complex inanimate objects and other areas in the parahippocampal gyrus and the lateral occipital lobe that showed more activity during the presentation of nonface objects. Event-related functional magnetic resonance imaging was then used to monitor activity in these extrastriate visual areas while subjects viewed Rubin's vase-face stimulus and indicated switches in perception. Since the spontaneous shifts in interpretation were too rapid for direct correlation with hemodynamic responses, each reported percept (faces or vase) was prolonged by suddenly adding subtle local contrast gradients (embossing) to one side or the other of the figure-ground boundary, stabilizing the percept. Under these conditions, only face-selective areas in the fusiform gyrus responded more strongly during the perception of faces. To control for effects of the physical change to Rubin's stimulus (i.e., addition of embossing), we compared activity when the face contours were embossed after the subject had just reported the onset of perception of either faces or vase. Activity in the fusiform face area responded more strongly under the first condition, despite the fact that the physical stimulus sequences were identical. Moreover, on a trial-to-trial basis, the activity was statistically predictive of the subjects' responses, suggesting that the conscious perception of faces could be made explicit in this extrastriate visual area.