SUMMARY The experiments reported here sought to investigate whether the K-complex evoked during sleep is comprised of activity from two separate physiological systems with different response properties. To that end, the parameters of stimulation in a two tone auditory ‘odd-ball’ task were varied systematically as stimuli were presented to subjects during NREM sleep. During experiment 1, the frequency (pitch) of the odd-ball stimulus varied systematically while intensity (loudness) was matched between tones. During experiment 2, pitch was matched between tones while the loudness of the odd-ball stimulus was varied. The long-latency event-related potentials (ERPs) N2 and P3 could be dissociated from the K-complex (N3 and P4) in response to these parametric manipulations. Information processing occurs during sleep, and is reflected in ERPs with a morphology largely analogous to those observed under similar conditions while subjects are awake. The second (K-complex) system is sleep specific. A model was constructed to explain the activity of these two hypothesized systems. As predicted by the model, K-complex latency was longer in Stage 2 when N2 and P3 were also active, than in Stage 4 where N2-P3 activity was lessened. These results support the two-system hypothesis; electrical brain activity evoked during sleep should not be considered a unitary sleep-specific response. Furthermore, the data indicate that the K-complex is sensitive to the physical characteristics of stimuli, that the sleeping brain processes information to a high degree, and that the ‘endogenous’ components of the ERP observed in awake humans reflect more automatic processes than previously suspected.