Eight normal young adult sleepers spent 4 nonconsecutive weeks in the laboratory. Each week consisted of a baseline night followed by 2 consecutive nights of disrupted sleep, followed by 2 recovery nights. Disruption conditions included: a) brief awakening after each minute of accumulated sleep, b) brief awakening after each 10 min of accumulated sleep, c) 2.5 hrs of normal sleep followed by a brief awakening at each sleep onset, and d) total sleep deprivation. Morning testing revealed that all disruption conditions decreased sleep latency in a morning nap test. Performance after 1-min disruptions approximated that seen after total sleep loss. Performance decrements were less in the 10-min condition and least in the 2.5-hr sleep condition. Performance under baseline and total sleep loss conditions was used to predict performance during the sleep deprivation condition using four sleep stage rules. Total time asleep and total time asleep minus stage 1 predicted performance poorly. Total SWS plus REM predicted performance best but could not differentiate the 10-min and 2.5-hr conditions. Therefore, it was concluded that the data were most parsimoniously explained by the Sleep Continuity Theory—i.e., that periods of uninterrupted sleep in excess of 10 min are required for sleep to be restorative.