Twelve subjects have been studied in a chamber that isolated them from external noise and lighting. After several control days, one group (n = 6) was subjected to 18 x 27-h 'days' and the other to 11 x 30-h 'days'. Sleep was in the dark, and awake times were spent in normal domestic lighting (150- 500 lux). Rectal temperature and wrist actimetry were measured throughout, and the phase of the circadian oscillator was inferred from that of the temperature data, purified to remove direct effects of activity. During the experimental 'days' the rhythms showed a mean period of 24.4 h. A detailed examination of the phase shifts from one day to the next showed that small advances and delays were superimposed upon this drift. Moreover, the mean size and direction of these shifts depended upon the time of exposure to lighting relative to the temperature minimum, as would be predicted from a phase-response curve.