Most previous studies of long-term maintenance therapy for bipolar illness simply compared proportions of patients experiencing recurrences. This method is inadequate because the length of time until a recurrence is not taken into account and because data from patients who withdraw prematurely from the study without a recurrence are either ignored or analyzed improperly. More appropriate survival analytic techniques were applied to the data from the National Institute of Mental Health Collaborative Study of bipolar patients treated with lithium carbonate, imipramine hydrochloride, or both. An interaction between treatment and the nature of the index episode (the episode that brought the patients into the study) was found. In patients with manic index episodes, both lithium and the combination were superior to imipramine. In patients with depressive index episodes, the combination was significantly superior to imipramine, whereas lithium was indistinguishable from imipramine. The latter finding differs from the original analysis, which found the combination to be no different from the other two treatments for patients with a depressive index episode.