Functional models of face processing have indicated that dissociations exist between the various processes involved, e.g. between familiar face recognition and matching of unfamiliar faces, and between familiar face recognition and facial expression analysis. These models have been successfully applied to the understanding of the different types of impairment that can exist in neuropsychological patients. In the present study, aspects of face processing in psychiatric patients were investigated in relation to Bruce & Young's (1986) model. Based on this functional model different predictions can be made. We contrast here the impaired expression analysis hypothesis, which is that psychiatric patients would show a deficit in facial expression recognition, but not in facial identity recognition or unfamiliar face matching, with the generalized deficit hypothesis, that patients would be impaired on all tasks. These hypotheses were examined using three forced‐choice tasks (facial recognition, facial expression recognition, and unfamiliar face matching) which were presented to schizophrenic and depressed patients, and to non‐patient controls. Results showed that schizophrenic patients performed at a significantly lower level than non‐patient controls on all three tasks, supporting the generalized deficit hypothesis.