In 2 experiments subjects were exposed for 3.5 weeks to a gradually (0.2 h/day initially) shortening day/night cycle, ending at 22.8 h and 22.0 h, respectively. Shortening of the cycle led to an initial but temporary increase of sleep latency. When the reduction ceased at 22.8 h and this length was maintained, sleep parameters were not further affected and the temperature rhythm in most subjects remained entrained to the 22.8 h period, although some instability occurred towards the end. In the 22.0 h experiment the continued reduction beyond 22.8 h led to disturbed sleep on day 15, at a day length of 22.4 h. Total sleep time, stage 2 and sleep efficiency were then markedly reduced. At this point sleep coincided with the peak of the body temperature rhythm and the amplitude of the latter was extremely small. This was also the point when the body temperature rhythm 'broke out' from the sleep/wake rhythm and showed a large 6 h phase jump (delay). Towards the end of the experiment, when sleep was initiated in the circadian temperature trough, REM propensity was increased. It was concluded that several sleep parameters were affected by the reduction of the day/night cycle although the specific effects depended on the amount of phase advance and on whether desynchronization occurred. Within the range of entrainment, however, most sleep parameters were remarkably unperturbed by the considerable changes of circadian parameters.