The role of extracorporeal membrane oxygenation (ECMO) in the treatment of the acute respiratory distress syndrome (ARDS) is controversial, notwithstanding the recent publication of the results of the CESAR (Conventional Ventilation or ECMO for Severe Adult Respiratory Failure) trial. Using Bayesian meta-analytic methods from three randomised controlled trials (RCTs) of ECMO in ARDS, we estimate the mortality odds ratio to be 0.78 (95% credible interval, 0.25-3.04), P (OR > 1) = 30%. Thus, a null effect of ECMO is not excluded and there appears only weak evidence of efficacy. We survey particular problems associated with the conduct of the "pragmatic" CESAR trial: composite endpoints, sample size estimation under uncertainty of baseline mortality rates, the generation of unbiased treatment comparisons, the impact of treatment non-compliance, and the uncertainty associated with cost-effectiveness and cost-utility analysis. We conclude that the CESAR trial is problematic in terms of both the clinical and economic outcomes, although observational series suggest plausible efficacy. We suggest that ECMO finds rationale as rescue therapy and that the current uncertainty of its role mandates a further RCT.