Nine healthy female subjects were studied when exposed to the natural light-dark cycle, but living for 17 'days' on a 27h day (9h sleep, 18h wake). Since the circadian endogenous oscillator cannot entrain to this imposed period, forced desynchronization between the sleep/activity cycle and the endogenous circadian temperature rhythm took place. This enabled the effects of activity on core temperature to be assessed at different endogenous circadian phases and at different stages of the sleep/activity cycle. Rectal temperature was measured at 6-minute intervals, and the activity of the nondominant wrist was summed at 1-minute intervals. Each waking span was divided into overlapping 3h sections, and each section was submitted to linear regression analysis between the rectal temperatures and the total activity in the previous 30 minutes. From this analysis were obtained the gradient (of the change in rectal temperature produced by a unit change in activity) and the intercept (the rectal temperature predicted when activity was zero). The gradients were subjected to a two-factor analysis of variance (ANOVA) (circadian phase/time awake). There was no significant effect of time awake, but circadian phase was highly significant statistically. Post hoc tests (Newman-Keuls) indicated that gradients around the temperature peak were significantly less than those around its trough. The intercepts formed a sinusoid that, for the group, showed a mesor (±SE) of 36.97 (±0.12) and amplitude (95% confidence interval) of 0.22°C (0.12°C, 0.32°C). We conclude that this is a further method for removing masking effects from circadian temperature rhythm data in order to assess its endogenous component, a method that can be used when subjects are able to live normally. We suggest also that the decreased effect of activity on temperature when the endogenous circadian rhythm and activity are at their peak will reduce the possibility of hyperthermia.