Résumé : BACKGROUND: Depressive subtypes generally have been neglected in research on treatment efficacy. We studied a sample of 699 severe unipolar depressed patients to detect any association between depressive features and treatment resistance. METHODS: Participants were divided into psychotic (PSY, n = 90), melancholic (MEL, n = 430) and non-melancholic (n = 179) subjects according to clinical features. Formal diagnostic criteria (Mini International Neuropsychiatric Interview items), and items from 17-item Hamilton Rating Scale for Depression (HRSD17) were compared across groups. Non-responders were defined by a HRSD17 cut-off score of !17 after the last adequate antidepressant treatment. Treatment-resistant depression (TRD) was defined as the failure to respond to !2 adequate antidepressant trials. Non-linear regression models were designed to detect associations between depressive subtypes and TRD. RESULTS: PSY and MEL patients appeared to be more severely affected and to share some "core" melancholic symptoms. Both PSY and MEL patients reported a higher rate of seasonality. However, we found no clinical or illness course variable associated with TRD. CONCLUSIONS: Our results indicate that psychotic and melancholic depression share some "core" melancholia symptoms, while no distinguishing psychopathological feature appears to be associated with TRD in severely depressed patients.