We examined ventilatory and waking responses to hyperoxic hypercapnia in 3 dogs during natural sleep. Progressive hypercapnia was induced by a rebreathing technique, and sleep was determined by electroencephalographic and behavioral criteria. In non-rapid eye movement sleep (high-voltage, slow-frequency electroencephalography) rebreathing continued for 0.99 +/- 0.05 min (mean +/- SE) before arousal occurred, and the alveolar PCO2, at arousal was 54.2 +/- 3.4 mm Hg. In contrast, during rapid eye movement sleep, rebreathing lasted for 1.71 +/- 0.27 min (P less than 0.05) before arousal occurred and the alveolar PCO2 at arousal was 60.3 +/- 4.2 mm Hg (P less than 0.05). Linear regression analysis of breath-by-breath instantaneous minute volume of ventilation, tidal volume, and respiratory frequency against alveolar PCO2 revealed regression coefficients in rapid eye movements sleep that were 14 to 33 per cent of those found in non-rapid eye movement sleep, and correlation coefficients of 0.26 to 0.46, compared to 0.71 to 0.91 in non-rapid eye movement sleep. Thus, the link between CO2 and ventilation appeared to be strong in non-rapid eye movement sleep but considerably disrupted during rapid eye movement sleep. We conclude that centers involved in both waking and ventilatory responses to hypercapnia behave as if they are less aware of or responsive to CO2 in rapid eye movement sleep than in non-rapid eye movement sleep.