OBJECTIVE This study was a 5-year naturalistic prospective follow-up of 54 consecutive admissions of adolescents to a university inpatient service with a diagnosis of bipolar I affective illness. METHODS Subjects received structured clinical evaluations every 6 months after entry to establish time to recovery and subsequent relapse. Regression models were used to identify predictors of differential course. RESULTS Rate of recovery varied by polarity of episode at time of entry, with quick recovery observed in subjects with pure mania or mixed states, and a protracted index episode in subjects with pure depression. Multiple relapses were most often seen in subjects with mixed or cycling episodes at intake. CONCLUSIONS Polarity of illness may have utility in identifying bipolar adolescents with a more recurrent illness. Comparison with adult data suggests that recurrence risks may vary as a function of age at onset or stage of the disease process.