The mismatch negativity (MMN) ERP component is generally considered to reflect the outcome of a pre-conscious change detection mechanism. The manipulation of active task demands has typically demonstrated that the MMN operates relatively independently of inferred attention. It remains a possibility, however, that subjects are capable of covertly sampling, or "eavesdropping" on, the irrelevant auditory stimuli, even during the most demanding of diversion tasks. The presence of the MMN in an unconscious state, such as natural sleep, provides strong evidence that its operations take place at a pre-conscious level. There exists consistent evidence that the MMN can be elicited at least during REM sleep, but these MMNs were typically elicited using oddball paradigms in which the new physical properties of deviants may trigger fresh afferent activation. The current sleep study employed a standard pattern in which two pure tones alternated (e.g. ABABABAB). Deviants were repetitions (e.g., ABABBBAB), and therefore physically identical to the preceding standard. In different conditions, the tones of the pattern were separated by either 1 or 6 semitones. A clear MMN was elicited in the waking state in the 6 semitone condition. The MMN was also elicited in the 6 semitone condition during REM sleep. No MMN was apparent in REM sleep in the 1 semitone condition. The MMN was not apparent in either the 6 or 1 semitone condition during NREM sleep. These results confirm the operation of the MMN in REM sleep, and support the view that the MMN operates at a pre-conscious level of processing.