Two hundred seventy severely hypoxemic (PaO2 < or = 55 mm Hg: mean +/- SD = 48 +/- 6) COPD patients (232 men) were selected for long-term oxygen therapy (LTOT). They were old (mean = 66 +/- 8 years), with severe airflow limitation (FEV1 = 30 +/- 12 percent of predicted), some CO2 retention (PaCO2 = 47 +/- 9 mm Hg), and compensated respiratory acidosis. Eighteen percent of the patients presented some complicating pleuropulmonary diseases (pleural thickening, sequelae of tuberculosis, etc). Overall survival proportion was poor: 70, 50, and 43 percent at 1, 2, and 3 years, respectively. The Cox model showed that the factors which independently reduced survival were lower CO transfer coefficient, smaller intrathoracic gas volume, more severe bronchial obstruction, the fact that oxygen administration did not increase PaO2 above 65 mm Hg, increasing age, and the presence of chest wall abnormalities. When the patients were divided into three groups according to mortality risk, the mean clinical and functional profile of the high-mortality risk group was consistent with the prevalence of emphysematous lesions. Moreover, the best survivors fitted better into the "bronchitic" type; they showed a higher mean PaCO2, suggesting that some degree of hypoventilation could delay muscular fatigue and improve survival. The difference in the proportion of "emphysematous" and "bronchitic" patients is a possible explanation for the variability of the mortality rate reported in literature.