Subjects in a time-free environment for 14 days estimated the hour and day several times a day. Half of the subjects were under a heavy exercise regime. During the waking hours, the no-exercise group showed no difference between estimated and real time, whereas the exercise group showed significantly shorter estimated than real time. Neither group showed a difference after the sleeping periods. However, the mean accumulated error for the two groups was 48.73 hours and was strongly related to the displacements of sleep/waking behavior. It is concluded that behavioral cues are the primary determinants of time estimates in time-free environments.