Continuous positive airway pressure (CPAP) is the most widely used therapy for obstructive sleep apnoea (OSA). Despite its general efficacy, oxygen desaturation due to hypoventilation persists in some patients. The present study analysed the factors which are associated with this primary failure and, moreover, examined the effect of a bilevel positive airway pressure (BiPAP) trial. In a 15-month period, 13 patients with OSA (Group A) failed to respond to initial CPAP therapy defined by a remaining apnoea/hypopnoea index (AHI) of > or = 5 or a mean nocturnal SaO2 < 90%. These patients were compared to an age- and AHI-matched control group (Group B) successfully treated by CPAP. A logistic regression analysis was performed to identify factors which are associated with initial failure to CPAP. Patients of the CPAP-failure group were treated with nasal BiPAP in the control mode. These patients were significantly more obese than patients of the control group (mean body mass index 44.2 +/- 7.7 vs 31.2 +/- 6.3 kg m-2; P < 0.001). PaO2 at rest (P < 0.001) and at exercise (P < 0.005) was significantly lower in Group A patients. PaCO2 at rest (P < 0.001) was significantly higher in Group A patients and changed for the worse during exercise, whereas it improved in the control group. Group A patients spent significantly (P < 0.0001) more time with oxygen saturation < 90%. The percentage of time spent at < 90% of SaO2 was the only factor which was independently associated with the initial failure of CPAP (OR 1.13; 95% CI 1.0-1.2). After 3 months of treatment with BiPAP, the patients' blood gas values while awake improved significantly (P < 0.05) for PaO2 as well as for PaCO2. In conclusion, patients with OSA resistant to initial CPAP are morbidly obese with impaired awake blood gas values. The percentage of time spent at < 90% of nocturnal SaO2 is independently associated with initial failure of CPAP. BiPAP in the control mode is adequate for nocturnal ventilation, and improves awake blood gas values.