Impaired recognition of facial emotion in schizophrenia is associated with poor social functioning. Evidence shows that targeted emotion recognition training (ERT) can improve perception of facial emotions in schizophrenia for up to 1 week after training. This study investigated whether (a) improved recognition generalizes to novel faces, (b) training effects are durable over 1 month, and (c) baseline functioning levels predict the extent of improvement. Thirty-nine participants with schizophrenia received ERT using Ekman's Micro-Expression Training Tool (METT; 2003). Emotion recognition was assessed using METT face stimuli and other face stimuli not used in training (static faces shown at 100% and 50% intensity and dynamic stimuli). Baseline ratings of interpersonal and cognitive functioning were collected; a subgroup of 10 participants was followed up at 1 month posttraining. Post-METT training, participants showed improved perception of METT faces and novel faces. The subgroup followed over 1 month showed improved recognition of novel faces and dynamic stimuli 1 month after training, but not immediately after training. Baseline measures of interpersonal and social functioning and general face processing and working memory abilities (50% intensity expressions only) predicted improvement in facial affect recognition immediately after METT training. These findings suggest that the effectiveness of ERT in schizophrenia is influenced by pretraining levels of social functioning and that general face processing abilities and working memory may affect the ability to accurately process subtle facial expressions. Furthermore, improved recognition generalizes to novel faces but only over time, which might indicate an increasing awareness of facial emotion after ERT, at least in people with better baseline social functioning.