Fifteen subjects who lived singly in an isolation unit without temporal cues were asked to note every day after awakening how many hours they thought they had slept, and in the evening before retiring how many hours they had been awake. These estimates of the duration of sleep and wakefulness were compared with the intervals between two signals given by the subjects by pressing a button at the time of waking up and when turning off their bedside reading lamp. The results can be summarized as follows: (a) the daily estimated durations of sleep and wakefulness were positively correlated with the actual durations in all but one subject; (b) sleep and wake times were better estimated in the presence of a light-dark cycle even if the subjects were not entrained by the zeitgeber; (c) for both episodes, there was a consistent trend from an overestimation of relatively short to an underestimation of long durations; (d) with equal durations in the two episodes, sleep was estimated to be shorter than wake time; (e) the most accurate estimates centered around 10.5 h of sleep and 13.5 h of wake time; (f) the sleep and wake times added up to 24 h in subjects who did not deliberately "compensate" for relatively long sleep estimates with a short wake estimate, with the full cycle being adjusted to 24 h.