The relationship between sleep and mood was examined in a longitudinal, naturalistic data set derived from out-patients with rapid-cycling bipolar disorder. Eleven patients completed daily self-ratings of mood and sleep logs for 18 months. Using logistic regression with autoregressive terms, we examined the effect of prior sleep (sleep duration, time of sleep onset, and time of wake onset) on the probability of being in a depressed, manic, or hypomanic episode on one or more subsequent days. Of the three sleep parameters, decreased sleep duration was the best predictor of mania or hypomania the next day, followed by wake onset time. The association between sleep duration and subsequent mood was less consistent for depression than for mania or hypomania. Four of the patients showed no relationship between mood and any of the sleep variables measured. These results reinforce the importance of monitoring, and perhaps controlling, sleep duration and wake onset time in at least some patients with rapid-cycling bipolar disorder.