Objectives: To determine the comparative effectiveness and cost-effectiveness of conventional ventilatory support versus extracorporeal membrane oxygenation (ECMO) for severe adult respiratory failure. Design: A multicentre, randomised controlled trial with two arms. Setting: The ECMO centre at Glenfield Hospital, Leicester, and approved conventional treatment centres and referring hospitals throughout the UK. Participants: Patients aged 18-65 years with severe, but potentially reversible, respiratory failure, defined as a Murray lung injury score ≥ 3.0, or uncompensated hypercapnoea with a pH < 7.20 despite optimal conventional treatment. Interventions: Participants were randomised to conventional management (CM) or to consideration of ECMO. Main outcome measures: The primary outcome measure was death or severe disability at 6 months. Secondary outcomes included a range of hospital indices: duration of ventilation, use of high frequency/oscillation/jet ventilation, use of nitric oxide, prone positioning, use of steroids, length of intensive care unit stay, and length of hospital stay - and (for ECMO patients only) mode (venovenous/veno-arterial), duration of ECMO, blood flow and sweep flow. Results: A total of 180 patients (90 in each arm) were randomised from 68 centres. Three patients in the conventional arm did not give permission to be followed up. Of the 90 patients randomised to the ECMO arm, 68 received that treatment. ECMO was not given to three patients who died prior to transfer, two who died in transit, 16 who improved with conventional treatment given by the ECMO team and one who required amputation and could not therefore be heparinised. Ninety patients entered the CM (control) arm, three patients later withdrew and refused followup (meaning that they were alive), leaving 87 patients for whom primary outcome measures were available. CM consisted of any treatment deemed appropriate by the patient's intensivist with the exception of extracorporeal gas exchange. No CM patients received ECMO, although one received a form of experimental extracorporeal arteriovenous carbon dioxide removal support (a clear protocol violation). Fewer patients in the ECMO arm than in the CM arm had died or were severely disabled 6 months after randomisation, [33/90 (36.7%) versus 46/87 (52.9%) respectively]. This equated to one extra survivor for every six patients treated. Only one patient (in the CM arm) was known to be severely disabled at 6 months. Patients allocated to ECMO incurred average total costs of £73,979 compared with £33,435 for those undergoing CM (UK prices, 2005). A lifetime model predicted the cost per quality-adjusted life-year (QALY) of ECMO to be £19,252 (95% confidence interval £7622 to £59,200) at a discount rate of 3.5%. Lifetime QALYs gained were10.75 for the ECMO group compared with 7.31 for the conventional group. Costs to patients and their relatives, including out of pocket and time costs, were higher for patients allocated to ECMO. Conclusions: Compared with CM, transferring adult patients with severe but potentially reversible respiratory failure to a single centre specialising in the treatment of severe respiratory failure for consideration of ECMO significantly increased survival without severe disability. Use of ECMO in this way is likely to be costeffective when compared with other technologies currently competing for health resources. Trial registration: Current Controlled Trials ISRCTN47279827.