In 320 patients with established bipolar I disorder, we examined the past course on the basis of polarity at onset (depressive, mixed, and manic). Despite the obvious limitations of retrospective methodology, information on course parameters in a large sample of affective disorders is most practically obtained by such methodology. We believe that our systematic interview of patients and their relatives--as well as the systematic study of their records--minimized potential biases. Depressive onsets were the most common, accounting for 50%, followed by mixed and manic onsets in about equal proportion. In general, the polarity of episodes over time reflected polarity at onset. Those with depressive onset had significantly higher levels of rapid cycling, as well as suicide attempts, but were significantly less likely to develop psychotic symptoms. Mixed onsets, too, had high rates of suicide attempts, but differed from depressive onsets in having significantly more chronicity yet negligible rates of rapid cycling at follow-up evaluation. Because cases with depressive onset had received significantly higher rates of psychopharmacologic treatment, our data are compatible with the hypothesis that antidepressants may play a role in the induction of rapid cycling. Overall, our data support the existence of distinct longitudinal patterns within bipolar I disorder, which in turn appear correlated with the polarity at onset. In particular, rapid cycling and mixed states emerge as distinct psychopathologic processes.