The purpose of this study was to determine whether a sleep log parameter could be used to estimate the circadian phase of normal, healthy, young adults who sleep at their normal times, and thus naturally have day-to-day variability in their times of sleep. Thus, we did not impose any restrictions on the sleep schedules of our subjects (n = 26). For 14 d, they completed daily sleep logs that were verified with wrist activity monitors. On day 14, salivary melatonin was sampled every 30 min in dim light from 19:00 to 07:30 h to determine the dim light melatonin onset (DLMO). Daily sleep parameters (onset, midpoint, and wake) were taken from sleep logs and averaged over the last 5, 7, and 14 d before determination of the DLMO. The mean DLMO was 22:48 +/- 01:30 h. Sleep onset and wake time averaged over the last 5 d were 01:44 +/- 01:41 and 08:44 +/- 01:26 h, respectively. The DLMO was significantly correlated with sleep onset, midpoint, and wake time, but was most strongly correlated with the mean midpoint of sleep from the last 5 d (r = 0.89). The DLMO predicted using the mean midpoint of sleep from the last 5 d was within 1 h of the DLMO determined from salivary melatonin for 92% of the subjects; in no case did the difference exceed 1.5 h. The correlation between the DLMO and the score on the morningness-eveningness questionnaire was significant but comparatively weak (r = -0.48). We conclude that the circadian phase of normal, healthy day-active young adults can be accurately predicted using sleep times recorded on sleep logs (and verified by actigraphy), even when the sleep schedules are irregular.