As constant hypoxaemia itself may trigger development of apnoea and periodic breathing, we have studied the effect of oxygen therapy on the occurrence of late postoperative episodic hypoxaemia. Thirty-five patients without cardiopulmonary disease and undergoing elective total hip replacement were monitored with a pulse oximeter on the second night after operation (23:00 to 07:00), receiving either 21% or 37% oxygen by face mask in a randomized double-blind design. Mean oxygen saturation was greater in the group receiving 37% oxygen than in those having 21% oxygen (96% vs 92%, P less than 0.01). There was a weak correlation between mean oxygen saturation and the total number of hypoxaemic episodes (rs = -0.62, P less than 0.001), explained partly by the calculated (non-mechanistic) reduction in mean saturation by the episodes of hypoxaemia. There was no significant difference between the groups in the total number of sudden decreases in oxygen saturation, the duration of the events or number of patients with events to more or less than 80% oxygen saturation, although there was a trend towards fewer patients having events to less than 80% in the 37% oxygen group (nine of 17 patients vs five of 18 patients (ns); 95% confidence limits of median difference: -6 to 56%). We conclude that postoperative oxygen therapy with 37% oxygen by face mask increases mean oxygen saturation, but does not influence the basic mechanism leading to episodic hypoxaemia.