Abstract PET was used to measure regional cerebral blood flow (rCBF) while memorizing pictures of unfamiliar human faces presented one at a time (FaceMemory). Other conditions included: (1) FaceRepeat—memorization of four individual faces presented repeatedly; (2) FaceWatching—viewing passively single faces without overt memory demands; and (3) Scrambled—counting dots superimposed on pictures of scrambled faces. After each FaceMemory condition and after the final FaceWatching condition scan, recall was tested by measuring face recognition. Contrasting FaceMemory and Scrambled conditions revealed several temporal activations: right midfusiform and bilateral anterior fusiform gyri. Contrasting FaceWatching and Scrambled conditions showed bilateral activation in the temporal poles and in the anterior fusiform gyri. No hippocampal activation arose from any contrast. Region of interest analyses on the above areas showed correlations with performance: (1) only rCBF in the right midfusiform correlated positively with encoding during the FaceMemory and FaceWatching conditions; (2) in the right temporal polar cortex rCBF decreased during FaceMemory and correlated positively with performance, whereas rCBF increased during FaceWatching and correlated negatively with incidental performance; and (3) activity in the anterior fusiform gyri remained constant across the conditions of FaceMemory, FaceRepeat, FaceWatching, and Scrambled and was uncorrelated with performance. These data suggest an expanded mnemonic role for the right midfusiform in depth of processing/encoding of face information, temporal polar cortex in face perception and recognition, and anterior fusiform activity in featural visual feature processing.