Emotion recognition is known to be impaired in schizophrenia patients. Although cognitive deficits and symptomatology have been associated with this impairment there are other patient characteristics, such as alexithymia, which have not been widely explored. Emotion recognition is normally assessed by means of photographs, although they do not reproduce the dynamism of human expressions. Our group has designed and validated a virtual reality (VR) task to assess and subsequently train schizophrenia patients. The present study uses this VR task to evaluate the impaired recognition of facial affect in patients with schizophrenia and to examine its association with cognitive deficit and the patients' inability to express feelings. Thirty clinically stabilized outpatients with a well-established diagnosis of schizophrenia or schizoaffective disorder were assessed in neuropsychological, symptomatic and affective domains. They then performed the facial emotion recognition task. Statistical analyses revealed no significant differences between the two presentation conditions (photographs and VR) in terms of overall errors made. However, anger and fear were easier to recognize in VR than in photographs. Moreover, strong correlations were found between psychopathology and the errors made.