The human circadian pacemaker, with an intrinsic period between 23.9 and 24.5 hr, can be reset by low levels of light. Biomathematical models of the human clock predict that light-dark cycles consisting of only approximately 3.5 lux during 16 hr of wakefulness and 0 lux during 8 hr of sleep should entrain approximately 45% of the population. However, under real-life conditions, sleep-wake schedules and the associated light-dark exposures are often irregular. It remains unclear whether the phase of the pacemaker would remain stable under such conditions. We investigated the stability of the circadian phase in dim light by assessing the plasma melatonin rhythm during nine consecutive circadian cycles. Ten subjects were scheduled to sleep for 8 hr (0.03 lux) and to be awake for 16 hr (5-13 lux) during all days except on days 4 and 8, during which the subjects were sleep deprived for 40 hr (5-13 lux), either in a sitting/standing or supine body posture. In all subjects, the phase of the melatonin rhythm occurred at a later clock time on day 9 than on day 2 (average delay: 1.4 hr). Largest delays in the melatonin onset were observed in subjects with low amplitude melatonin rhythms. The area under the curve during active melatonin secretion was significantly reduced when subjects were sleep deprived in the 40-hr supine body posture condition compared with either the 40-hr sitting/standing sleep deprivation (SD) or the ambulatory condition under non-SD conditions. Posture differences did not significantly affect the relative phase position of the melatonin profiles. The data indicate that under conditions of reduced zeitgeber strength, the phase of the human circadian pacemaker, using plasma melatonin as a marker, can be phase delayed by one night of SD and the associated dim light exposure.