Prosopagnosic patients appear to be impaired at recognizing faces. However, recent evidence for “covert recognition” in prosopagnosia has been taken to suggest that the impairment is not in face recognition per se, but rather in conscious access to face recognition. The most widely used test for covert recognition of faces in prosopagnosia is the face-name relearning task, in which some prosopagnosics have been found to learn correct names for previously familiar faces more easily than incorrect names. Although this phenomenon is consistent with face recognition operating normally but out of reach of conscious awareness, it may also be consistent with an impairment in face recognition per se. Perhaps savings in relearning is sufficiently sensitive to the residual information contained in degraded face representations that are not detectable by overt measures of recognition. If so, then we should expect to observe this same savings in relearning when overt recognition is obliterated for reasons other than brain damage. In the present study, we used forgetting of face-name associations in normal subjects as a way of degrading recognition ability. We found the same dissociation between overt recognition performance and savings in relearning as observed in prosopagnosic patients. This implies that the performance of prosopagnosic patients in these tasks does not demand explanation in terms other than an impairment in face recognition per se.