Polygraphic recordings were performed in 14 sleeping premature infants receiving ventilation for respiratory distress syndrome. All were clinically stabilized, with normal EEG and neurologic status and differentiated sleep states (coded according to EEG and REM criteria). They all had two respiratory patterns: passive, completely dependent on the ventilator, and active, with autonomous respiratory movements and/or inspiratory diaphragmatic activity added to passive respiration. We found that in infants ventilated at the rate of 18-54/min, respiration was more active and autonomous in active REM sleep and more passive and dependent on the machine in quiet NREM sleep (P less than 0.005). Within the limits of the values observed in our study, differences between sleep states were not due to other factors that could possibly interfere with and modify the degree of respiratory autonomy. We found no significant correlation between the percentage of time passed with active respiration on one hand and age (gestational, postnatal, conceptional) or diagnostic or physical parameters of artificial ventilation and blood gas levels on the other hand. Our results suggest that in artificially ventilated but neurologically normal premature infants, differences between respiratory control in both sleep states exist as early as 28 weeks conceptional age (lower limit of our study).