Extracorporeal membrane oxygenation (ECMO) is a complex and expensive technology that can be used to provide temporary support during respiratory failure. Its value for mature newborn infants is controversial because of varying interpretation of the available evidence. We undertook a collaborative trial throughout the UK to assess whether a policy of referral for ECMO has a beneficial effect on survival to 1 year without severe disability in comparison with conventional management. Between 1993 and 1995, 185 mature (gestational age at birth > or = 35 weeks, birthweight > or = 2 kg) newborn infants with severe respiratory failure (oxygenation index > or = 40) were enrolled from 55 hospitals in a randomised comparison of either referral to one of five specialist centres for consideration of ECMO or continued intensive conventional management at the original hospital. The most common diagnoses were persistent pulmonary hypertension due to meconium aspiration, congenital diaphragmatic hernia, isolated persistent fetal circulation, sepsis, and idiopathic respiratory distress syndrome. Of the infants allocated ECMO, 84% received this support. Recruitment to the trial was stopped early (November, 1995) by the trial steering committee on the advice of the independent data-monitoring committee, because the data accumulated showed a clear advantage with ECMO. 124 children were enrolled before December, 1994; those who survived to 1 year of age have undergone neurological assessment at that age (by one of three developmental paediatricians unaware of treatment allocation). Overall, 81 (44%) infants died before leaving hospital, and two are known to have died later. Death rates differed between the two trial groups; 30 of 93 infants allocated ECMO died compared with 54 of 92 allocated conventional care. The relative risk was 0.55 (95% Cl 0.39-0.77; p = 0.0005), which is equivalent to one extra survivor for every three to four infants allocated ECMO. The difference in survival applied irrespective of the primary diagnosis, disease severity, and type of referral centre. The benefit of ECMO was also found for the primary outcome measure of death or disability at 1 year (among 124 children enrolled before December, 1994). One child in each group has severe disability (overall Griffiths' developmental quotient < 50, or untestable), and 16 (ten ECMO, six conventional management) have impairments with a lesser degree of disability. These preliminary results demonstrate the clinical effectiveness of a well-staffed and organised neonatal ECMO service. ECMO support should be actively considered for neonates with severe but potentially reversible respiratory failure.