We assessed improvement patterns and predictors of outcome over a 1-year period, in a sample of depressed patients receiving treatment from a specialized mood disorders unit. Patients with melancholia had a differential improvement pattern from the nonmelancholics in the first 20 weeks, but case rates and severity levels were comparable at 20 weeks and at 1 year. Only three variables (older age at first episode, less severe depression and extraversion) were predictors of improvement in both groups. Improvement was predicted by less psychomotor disturbance, absence of personality disorder, and higher social functioning in the melancholic patients. A reported absence of timidity and shyness in childhood, a briefer duration of depression, and receipt of individual psychotherapy predicted a better outcome in the nonmelancholic patients. Although significant predictors were few overall, the suggested differential relevance for most of the isolated predictors argues for outcome studies that examine melancholic and nonmelancholic depressive disorders separately.