We found significant differences in time to recovery and rates of chronicity in 155 patients with bipolar illness when the episodes were subtyped into those with manic symptoms alone (pure manic), depressive symptoms alone (pure depressed), or symptoms of depression and mania (mixed or cycling) up to the time of entry into a clinical research study. Most of the patients in all three groups who did not recover received levels of somatotherapy that were generally consistent with current recommendations for intensity of treatment appropriate to each condition. Based on a median follow-up of 18 months, the life-table estimate of the probability of remaining ill for at least one year was 7% for the pure manic patients compared with 32% in patients who entered the study with episodes that were mixed or cycling. Purely depressed patients had a 22% probability of remaining ill, approximating rates found in patients without bipolar illness who have episodes of depression. Different clinical variables were found to predict time to recovery in each of these groups. We propose that this subtyping of episodes may be a clinically useful part of the classification of bipolar disorders.