All night sleep/wake EEGs were examined for diagnostic features sensitive to early Alzheimer's disease (AD) using computer automated techniques. Thirty-nine mild AD patients and 43 normal controls underwent 9 h of EEG recording in the sleep laboratory. All-night EEGs were screened for ideal, low artifact tonic REM sleep using autoregressive and power spectral techniques. The frequency spectra during tonic REM sleep revealed a significant shift towards slower wave forms in AD vs. control subjects. Beta (greater than 12 Hz) was reduced and theta and delta (2-8 Hz) increased in AD compared to control groups. This frequency shift was demonstrated by several analytic techniques, including binned spectral energies and unique zones in the frequency spectra. Discriminant analyses using optimal binned EEG variables correctly classified 74% of AD and 98% of control subjects, and unique zone scores correctly classified 92% of AD and 95% of control subjects, indicating that these sleep EEG changes are apparently predictive of AD status.