Accuracy at perceiving frontal eye gaze was studied in monkeys and human subjects using a forced-choice detection task on paired photographs of a single human face. Monkeys learned the task readily, but after bilateral removal of the banks and floor of the superior temporal sulcus (STS) they failed to perform the task efficiently. This result is consistent with the conclusion, based on recordings from single cells in awake, behaving monkeys [Perret et al., Physiological Aspects of Clinical Neuro-ophthalmology, Chapman & Hall, London, 1988] that this region of the temporal lobe is important for coding information about eye-gaze of a confronting animal. Human subjects were given identical stimuli in a task where they were asked to detect "the face that is looking straight at you". Human performance is sensitive to the degree of angular deviation from the frontal gaze position, being poorest at small angular deviations from 0 degrees. This was also true of monkeys viewing these stimuli, pre- and post-operatively. Compared with normal controls, two humans prosopagnosics were impaired at this task. However the extent of impairment was different in the two patients. These findings are related to earlier reports (including those for patients with right-hemisphere damage without prosopagnosia), to normal performance with upright and inverted face photographs, and to notions of independent subsystems in face processing.