Assessment of different hospitals' performance in neonatal intensive care has tended to rely on risk of mortality adjusted only for birthweight. We have developed a neonatal scoring system, CRIB (clinical risk index for babies), to take account of other factors. Scores are given for birthweight, gestational age, maximum and minimum fraction of inspired oxygen and maximum base excess during the first 12 h, and presence of congenital malformations. CRIB was developed retrospectively in a cohort of 812 infants of birthweight 1500 g or less or gestational age less than 31 weeks treated in four UK tertiary hospitals between 1988 and 1990. The scoring system was then validated by comparing its value as a predictor of hospital death with that of birthweight in a separate cohort of 488 similar infants. The area under the receiver operating characteristic (ROC) curve for predicting death in this validation cohort (a measure of the predictor's accuracy) was significantly greater for CRIB than for birthweight alone (0.90 [SE 0.05] vs 0.78 [0.03], p = 0.03). Both indices of initial risk were used to assess the performance of nine tertiary and four non-tertiary hospitals in the UK. We looked at hospital mortality among the 1198 infants in the development and validation cohorts plus 350 other infants of birthweight 1500 g or less or gestational age less than 31 weeks [corrected]. Without adjustment for initial risk, there was no clear difference in group mortality between the tertiary and non-tertiary hospitals. However, calculation of risk-adjusted mortality by means of CRIB showed that babies were twice as likely to die in the non-tertiary hospitals as in the tertiary hospitals (odds ratio 2.12 [95% CI 1.39-3.24]). Adjustment for birthweight alone also showed higher odds of death in non-tertiary hospitals (1.45 [1.01-2.11]). CRIB is a robust index of initial neonatal risk that is more accurate than birthweight and simple enough for routine use. International comparisons of staffing policy and organisation of neonatal intensive care units are needed to investigate the reasons for variations in performance.