Benzodiazepines can shift the phase of circadian rhythms in mammalian species, but few data are available on their phase-response effects in humans, and on possible links between timing of administration and hypnotic efficacy. Using a placebo-controlled, cross-over design, we evaluated the hypnotic effect of lormetazepam 0.03 mg/kg and placebo in 38 inpatients who were affected by a major depressive episode. Patients were divided into three groups, receiving treatment at 18.00 h, 20.00 h or 22.00 h, respectively. Sleep and psychiatric symptoms were evaluated with self-administered scales and a sleep diary. The results demonstrate that active treatment significantly improved insomnia independent of the severity of depression, which remained unchanged. Timing of treatment influenced changes in timing of sleep observed with active treatment. Although sleep duration was equally improved in all treatment groups, patients who received treatment at 20.00 h showed an acute advance of sleep onset, with no changes in morning awakening. Patients who received treatment at 22.00 h showed an acute delay in morning awakening, with no changes of sleep onset. Finally, patients who received treatment at 18.00 h showed a non-significant trend in the same direction. These effects reverted with cross-over return to placebo. The perceived degree of improvement of insomnia was proportional to the advance in timing of sleep onset obtained with treatment. Our results suggest that the effects of lormetazepam on the subjective sleep of patients affected by a major depressive episode depend upon the timing of administration, and that improvement in subjective sleep is related to advance of sleep onset, and not to delay of morning awakening.