Pollack et al reported a trend to improved risk adjusted mortality in US Paediatric Intensive Care Units (PICUs) after junior residents' first 3 months in post (JAMA 1994; 272: 941-6). We tested a similar hypothesis in neonatal intensive care units (NICUs), after adjusting for initial risk of death using CRIB (clinical risk index for babies: Lancet 1993; 342: 193-8). This was a retrospective cohort study of 9,579 infants of birth weight <1500g or gestation <31 weeks in 1992-4 from self-selected NICUs which returned questionnaires on staffing policy and organisation. Expected mortality was calculated by summing the individual probabilities of death for infants admitted to each NICU using a previously validated multiple logistic regression equation based on CRIB. Data on CRIB, junior doctors' starting dates and hospital mortality are reported for 8,820 (92%) infants from 67 NICUs in 19 countries. On logistic regression, risk adjusted mortality was lower after junior doctors' first 3 months: Odds Ratio (OR) 0.85, 95% Confidence Interval (CI) 0.75-0.98. Improvement after the first 3 months was greatest in Australia (5 NICUs, n = 1,501): OR 0.65, 95% CI 0.44-0.96. No improvement was seen in the UK (34 NICUs, n = 3,531): OR 1.02, 95% CI 0.95-1.09. For the whole period, the standardised ratio of observed to expected mortality (SMR) was lower in the Australian NICUs, SMR 0.68, 95% CI 0.79-0.91, than in the UK NICUs, SMR 1.02, 95% CI 0.96-1.09. This study supports the hypothesis that mortality may fall after junior doctors' first 3 months. A secondary finding was that risk adjusted mortality was lower in the Australian than UK NICUs. These results have major implications and need to be replicated in larger, nationally representative cohorts in a prospective collaboration. Studies of training, supervision, organisation and risk adjusted outcomes in NICUs and PICUs are also indicated.