Ventilation and gas exchange were studied during sleep and incremental treadmill exercise in 19 patients with severe stable COPD with the primary aim of comparing the pathophysiology of oxygen desaturation in the two conditions. A secondary aim was to determine whether exercise studies could aid in the prediction of sleep desaturation. Full polysomnography was used, and ventilation, arterial oxygen saturation (SaO2), and transcutaneous PCO2 (PtcCO2) were monitored continuously during sleep. No patient had significant sleep apnea. Mean (SD) FEV1 was 32 (9.1)% predicted, PaO2 was 71.2 (12.4) mm Hg, and PaCO2 was 44.5 (4.6) mm Hg. SaO2 fell twice as much during sleep as during maximum exercise: 13.1 (8.9) vs 6.0 (3.6)% (p < 0.001). The mean sleep and exercise SaO2, and minimum sleep and exercise SaO2 were well correlated on linear regression (r = 0.81 and 0.78, respectively, p < 0.001), but on multiple regression analysis, awake PaO2 was a better predictor of sleep desaturation than was exercise desaturation. The 12 major desaturators (minimum sleep SaO2 < 85%) had twice as great a fall in exercise SaO2 as the 7 minor desaturators (3.6 +/- 2.8 vs 7.4 +/- 3.3%, p < 0.05). The major desaturators also had a greater fall in estimated sleep PaO2: 19.8 (5.1) vs 6.4 (7.1) mm Hg (p < 0.01), which suggests that their greater sleep desaturation is not simply due to their position on the steep portion of the oxyhemoglobin dissociation curve. The rise in PtcCO2 during sleep was similar among major and minor desaturators: 7.5 (2.9) vs 5.8 (3.7) mm Hg (p = NS), suggesting that all patients had a similar degree of hypoventilation during sleep, and that the greater fall in SaO2 and estimated PaO2 among some patients was secondary to other factors such as increased ventilation-perfusion mismatching.