For some time, it has remained uncertain whether the circadian rhythms of permanent night shift workers are adapted to their night-active schedule. Previous studies of this question have often been limited by "masking" (evoked) effects of sleep and activity on body temperature and cortisol, used as marker rhythms. In this study, the problem of masking was minimized by measuring the timing of melatonin production under dim light conditions. Nine permanent night shift workers were admitted to the Clinical Research Center (CRC) directly from their last work shift of the week and remained in dim light while blood samples were obtained hourly for 24 hours. Melatonin concentrations were measured in these samples using a gas-chromatographic mass-spectrometric method. Sleep diaries were completed for two weeks prior to the admission to the CRC. Overall, the onset of the melatonin rhythm was about 7.2 hours earlier (or 16.8 hours later) in the night workers compared to day-active controls. It was not possible to know whether the phase of the melatonin rhythm was the result of advances or delays. In night shift workers, sleep was initiated (on average) about three hours prior to the onset of melatonin production. In contrast, day-active subjects initiated sleep (on average) about three hours after their melatonin onset. Thus, the sleep times selected by night shift workers may not be well-synchronized to their melatonin rhythm, assumed to mark the phase of their underlying circadian pacemaker.