Polysomnography was performed during two consecutive nights in 23 patients with major depression. After every final awakening patients estimated the change of their mood from evening to morning: 1. Mood worse in the morning than in the evening; 2. Mood does not change; 3. Mood better in the morning. When mood was estimated as being better in the morning (20% of all nights), eye movement density in REM sleep increased from the first to the fourth cycle. In all other nights eye movement density was slightly higher in the first than in subsequent cycles. Mood improvement correlated positively with eye movement density in the fourth cycle and negatively with REM sleep duration in the first cycle. Eye movement density in the first cycle correlated positively with the subjective estimation of sleep latency. A possible functional difference between initial and final REM sleep periods is proposed.