Perceiving faces is critical for social interaction. Evidence suggests that different neural pathways may be responsible for processing face identity and expression information. By using functional magnetic resonance imaging (fMRI), we measured brain responses when observers viewed neutral, fearful, and scrambled faces, either visible or rendered invisible through interocular suppression. The right fusiform face area (FFA), the right superior temporal sulcus (STS), and the amygdala responded strongly to visible faces. However, when face images became invisible, activity in FFA to both neutral and fearful faces was much reduced, although still measurable; activity in the STS was robust only to invisible fearful faces but not to neutral faces. Activity in the amygdala was equally strong in both the visible and invisible conditions to fearful faces but much weaker in the invisible condition for the neutral faces. In the invisible condition, amygdala activity was highly correlated with that of the STS but not with FFA. The results in the invisible condition support the existence of dissociable neural systems specialized for processing facial identity and expression information. When images are invisible, cortical responses may reflect primarily feed-forward visual-information processing and thus allow us to reveal the distinct functions of FFA and STS.