1. Sixteen rats were recorded continuously for 3 days using an automated system that detected, quantified, and stored the incidence of cortical delta waves, cortical sigma spindles, hippocampal theta rhythm, and electromyographic activity. A time series then was constructed wherein 15-s epochs were ascribed to one behavioral state: wakefulness (W), quiet sleep (QS), or active sleep (AS, a state also referred to as REM sleep). From those series, AS episodes and non-AS intervals could be determined. Episodes and intervals were defined as lasting at least two epochs and the one-epoch episodes and intervals were incorporated to the ongoing state. 2. Having established the length of each AS episode and non-AS interval, pairings were made, on the one hand between episodes and their preceding intervals, and on the other, between episodes and the intervals that followed. 3. Highly significant correlations were found between the length of AS episodes and the length of the non-AS intervals that followed. Correlations were also significant when calculated separately versus the amount of QS and of W within the following interval. Correlations improved when they were performed against the log of the interval and when only intervals with a predominance of QS were selected. 4. No significant correlation was found between the length of AS episodes and the length of the preceding non-AS intervals, except for a negative one that was present only when the statistical analysis was performed in the unsmoothed array where the one-epoch episodes and intervals were preserved. 5. These results suggest that there is a short- term homeostasis operating within the spontaneous architecture of sleep in rats. This homeostatic mechanism is not manifested by the regulation of the length of AS episodes. Instead, there is a forward regulatory mechanism that, given the duration of an AS episode, permissively controls the interval that the animal may abstain from AS, and hence the timing of the triggering of a new AS episode.