Twelve healthy participants were studied in an Isolation Unit. For the first 7 (control) days, subjects lived on UK time. Then the clock was advanced by 8 h, mimicking an eastward time-zone transition, and for days 8 to 12, participants lived on this new local time. Two constant routines (participants were not allowed to sleep, were restricted in movement, and ate regular, identical snacks) were undertaken, during the control days (days 3 to 4) and at the end of the experiment (days 11 to 12). Rectal temperature and activity were measured throughout, with activity used to correct the measured temperatures for the direct (masking) effects of the sleep-wake cycle. Phase changes of the temperature rhythm between the constant routines were assessed by cross-correlation and cosinor analysis. During days 8 to 10, the measured temperatures and those that had been corrected (purified) for masking were assessed by the same two methods, and the shifts were extrapolated to predict the values expected during the second constant routine. Individuals differed widely in the phase shifts of the temperature rhythm, but the correlations between the changes measured by constant routines and those estimated by the purification methods were high (r=0.771 to 0.903), and the differences between them were not significantly different from zero (p>0.24). Phase shifts of the measured (masked) temperature rhythm were poorer predictors of the shift obtained from the constant routines (r<or=0.605; mean+/-SD of differences>3+/-4.5 h). Limitations of the methods due to the variability of results are discussed, but we conclude that the mean phase shifts obtained from purified, but not raw, temperature data show acceptable agreement with those found using our version of the constant routine.