We measured ventilatory and arousal responses to progressive eucapnic hypoxia during wakefulness, nonrapid-eye-movement (NREM) sleep, and rapid-eye-movement (REM) sleep using a progressive isocapnic rebreathing method. Nine healthy adults (4 female, 5 male) slept with a mask glued to the face with medical silicone rubber and breathed from a closed valveless biased flow circuit, including an in-line bag-in-box and a variable soda-lime absorber. Progressive hypoxia was induced by consumption of oxygen and by gradual replacement of circuit volume with nitrogen. Tidal volume was measured by electrical integration of the flow signal from a pneumotach on the box. Arterial hemoglobin oxygen saturation (SaO2) was measured with an ear oximeter and end-tidal CO2 tension (PetCO2) was measured continuously and kept constant by variable absorption. Sleep state was identified using standard criteria with 2 channels each of EEG, submental EMG, and EOG. There was marked variability in arousal level both in NREM and REM sleep, with subjects failing to awaken by 70% SaO2, our previously agreed safety limit, on 12 of 26 NREM tests, and 7 of 15 REM tests. During wakefulness, the mean slope +/- SEM of the ventilatory response to hypoxia was 0.68 +/- 0.07 L/min% SaO2 (n = 36, mean PetCO2 = 37.0 mmHg). In NREM sleep, this response decreased to a mean of 0.42 +/- 0.06 L/min/% SaO2 (n = 26, mean PetCO2 = 37.2 mmHg). In REM sleep, the average ventilatory response was further decreased to 0.33 +/- 0.06 L/min/% SaO2 (n = 15, mean PetCO2 = 37.8 mmHg). Analysis of variance showed a significant state-dependent effect on ventilatory response (p less than 0.01). The wake-NREM and wake-REM differences were significantly different (p less than 0.05), but the NREM-REM difference was not (p greater than 0.2). In REM sleep, breath-to-breath variability was marked, and in 2 cases, the response was not significantly different from zero. In all 3 states, the entire ventilatory response was due to increments in tidal volume. We conclude that (1) at normal alveolar CO2 tension, hypoxia is a poor arousal stimulus in humans, both in NREM and REM sleep, and (2) the eucapnic hypoxic response is reduced but present in NREM sleep and similarly reduced but not always present in REM sleep.