Sixty adolescents, aged 13 to 16 years, hospitalized for major depression were studied prospectively, for three to four years to determine the utility of clinical, genetic, and pharmacologic response variables in predicting a bipolar course of illness. Bipolar outcome was observed in 20% of the cohort. Statistical analyses showed that bipolarity was predicted by (1) a depressive symptom cluster comprising rapid symptom onset, psychomotor retardation, and mood-congruent psychotic features; (2) a "loading" of affective disorder in the family pedigree, a family history of bipolar illness, and the presence of illness in three successive generations of the pedigree; and (3) pharmacologically induced hypomania. All predictors were shown to have high specificity for bipolar outcome, whereas pharmacologic hypomania and symptom cluster permitted the highest confidence of prediction, 100% and 80%, respectively. Even in juvenile depression, careful attention to clinical and biologic variables may aid in the predictive differentiation of meaningful diagnostic subtypes.