Seventy-one patients scheduled to undergo upper or lower abdominal surgical procedures were allocated at random to one of seven treatment groups: in the recovery room they were to receive oxygen via a 40% Ventimask with 10 litres/minute oxygen flow, or via either a Hudson mask or a nasal cannula with 3, 6 or 9 litres/minute oxygen flow. The 40% Ventimask gave the most consistent, satisfactory postoperative values of PaO2 but the much cheaper nasal cannula at 6 or 9 litres/minute was generally adequate in conscious patients. The performance of the intermediately priced Hudson mask was similar to that of the nasal cannula at these flows. The unconscious state was associated with a 45% lower PaO2 than the rousable or awake states. Differences between the treatments with regard to postoperative PaCO2 were small and non-significant. The nasal cannula with 6 litres/minute humidified oxygen flow is recommended for routine treatment, and the Ventimask for unconscious patients.