Human visual evoked potentials were recorded during presentation of photos of human and animal faces and various face features. Negative waves with approximate peak latencies of 165 msec (N170) were bilaterally recorded from the occipito-temporal regions. Mean peak latencies of the N170 were shorter for faces than eyes only. Analyses of amplitudes of evoked potentials indicated that the N170 elicited by faces reflected activity of a specific neural system which was insensitive to detailed differences among individual faces regardless of species, and consequently suggest that this system might function to detect existence of faces in general. On the other hand, the mean amplitude of the N170 elicited by human eyes was significantly larger than those by animal eyes. These differences in response latencies and amplitudes of the N170 suggest existence of at least 2 different visual evoked potentials with similar latencies (i.e., N170) which are sensitive to faces in general and human eyes, respectively. Dipole source localization analysis indicated that dipoles for the N170 elicited by eyes were located in the posterior inferior temporal gyrus, and those for faces, located initially in the same region, but moved toward the fusiform and lingual gyri at the late phase of the N170. The results indicated that information processing of faces and eyes separated at least as early as the latency of the N170 at the posterior inferior temporal gyrus as well as the fusiform and lingual gyri, and might provide neurophysiological and anatomical bases to an initial structural encoding stage of human faces.