Forty-four normal male students, aged 18-23 years were studied. After adaptation and baseline night (BN), 3 or 4 consecutive nights were interrupted by a forced awakening (10-90 min) once a night (ENs). Subjects (Ss) were awakened after they had slept for the first sleep cycle plus 20 min of NREM sleep in the 2nd cycle. The REM latencies following return to sleep showed a bimodal distribution separated by 25-30 min. The ENs were divided into 2 clusters: SOREMP (sleep onset REM period) and non-SOREMP nights. After interruption, the 2nd and the 3rd REM durations increased on non-SOREMP nights compared to SOREMP nights. We plotted, separately for SOREMP and non-SOREMP nights, the fluctuation of REM episode probability (FRP) at successive points in time. We examined the correspondence of FRPs derived from the sleep-independent, the sleep-dependent, and the reset hypotheses, with FRP of intact BN. On both SOREMP and non-SOREMP nights, none of the 3 models corresponded with BN. Thus, we suggest that, for both SOREMP and non-SOREMP nights, intervening wakefulness cancels the pre-awakening REM rhythm, and a new REM rhythm starts with or without SOREMP. We discuss factors influencing the rate of SOREMP occurrence (SOREMP %), such as circadian effect, individual differences, length of interruption, and pre-awakening NREM duration.