Activation in or near the fusiform gyrus was estimated to faces and control stimuli. Activation peaked at 165 ms and was strongest to digitized photographs of human faces, regardless of whether they were presented in color or grayscale, suggesting that face- and color-specific areas are functionally separate. Schematic sketche evoked approximately 30% less activation than did face photographs. Scrambling the locations of facial features reduced the response by approximately 25% in either hemisphere, suggesting that configurational versus analytic processing is not lateralized at this latency. Animal faces evoked approximately 50% less activity, and common objects, animal bodies or sensory controls evoked approximately 80% less activity than human faces. The (small) responses evoked by meaningless control images were stronger when they included surfaces and shading, suggesting that the fusiform gyrus may use these features in constructing its face-specific response. Putative fusiform activation was not significantly related to stimulus repetition, gender or emotional expression. A midline occipital source significantly distinguished between faces and control images as early as 110 ms, but was more sensitive to sensory qualities. This source significantly distinguished happy and sad faces from those with neutral expressions. We conclude that the fusiform gyrus may selectively encode faces at 165 ms, transforming sensory input for further processing.