Cardiorespiratory behavior during sleep has been investigated by comparing visually analyzed minutes of EEG sleep with the digitized values of these two physiologic variables for each corresponding minute. Continuous 3-h nighttime sleep studies on 37 full-term and preterm neonates at comparable postconceptional term ages were acquired under controlled conditions, using a 24-channel computerized monitoring system and an automated event-marker program. Five thousand, two hundred ninety-four minutes were assigned an EEG state by traditional criteria. Eighteen preterm infants were compared with 19 full-term infants with respect to six cardiac and six respiratory measures: two nonspectral calculations (i.e. average per minute and variance of the means) and four spectral calculations of the cardiorespiratory signal (i.e. bandwidth, spectral edge, mean frequency, and ratio of harmonics). The relative capabilities of these measures to predict a sleep state change were investigated using discriminant analysis. A stepwise selection algorithm in discriminant analysis was used to identify the order of significance for the remaining variables. Eight cardiorespiratory measures were then submitted to multivariate analysis of variance to assess sleep state or preterm-full-term differences: mean frequency, bandwidth, average per minute, and ratio of harmonics for cardiac signals; and spectral edge, mean frequency, logarithm of variance, and ratio of harmonics for respiratory signals. Differences among the sleep states and between neonatal groups were highly significant (p < 0.0001). Interaction between sleep state and neonatal group was also significant (p < 0.034). Two variables differentiated preterm from full-term respiratory behavior: ratio (p < or = 0.001) and mean frequency (p < or = 0.02).(ABSTRACT TRUNCATED AT 250 WORDS)