Objectives: Use of appropriate face processing strategies is important for facial emotion recognition, which is known to be impaired in schizophrenia (SZ) and bipolar disorder (BD). There is preliminary evidence of abnormalities in the use of face processing strategies in the former, but there has been no explicit attempt to assess face processing in patients with BD. Methods: Twenty-eight BD I, 28 SZ, and 28 healthy control participants completed tasks assessing featural and configural face processing. The facial inversion effect was used as a proxy of second order configural face processing and compared to featural face processing performance (which is known to be relatively less affected by facial inversion). Results: Controls demonstrated the usual second-order inversion pattern. In the BD group, the absence of a second-order configural inversion effect in the presence of a disproportionate bias toward a featural inversion effect was evident. Despite reduced accuracy performance in the SZ group compared to controls, this group unexpectedly showed a normal second-order configural accuracy inversion pattern. This was in the context of a reverse inversion effect for response latency, suggesting a speed-versus-accuracy trade-off. Conclusions: To our knowledge, this is the first study to explicitly examine and contrast face processing in BD and SZ. Our findings indicate a generalized impairment on face processing tasks in SZ, and the presence of a second-order configural face processing impairment in BD. It is possible that these face processing impairments represent a catalyst for the facial emotion recognition deficits that are commonly reported in the literature. (JINS, 2016, 22, 652–661)