Although there has been a considerable amount of research into sleep using either subjective estimates or objective (EEG) estimates, there has been no attempt to validate the accuracy of subjective estimates of sleep. The present study investigates the relation between subjective estimates of total sleep time, delay to sleep onset and the number of awakenings in eight normal subjects over 116 subject-nights. They were studied before taking drugs, during drug administration and in withdrawal. It was found that they underestimated total sleep time but overestimated the delay to sleep onset and the number of awakenings. There was a significant withdrawal effect. The implications of this withdrawal effect are discussed in relation to the design of hypnotic drug trials. The reliability of norms for sleep are also discussed.