Mortality in the first 2 years of 634 very-low-birthweight infants admitted to eight neonatal intensive care units in Italy, and the factors associated with the net probability of death from each cause, were studied by means of the Cox proportional hazard model. A clinical classification of the causes of death was used. Overall mortality was 33.7% (intercentre range 12.6-52.9%). The highest cause-specific mortality rates were observed for respiratory problems, intra-ventricular haemorrhage (IVH) and infections (14.5%, 6.3% and 5.7% respectively). The leading causes of death were respiratory problems and IVH in the first week of life, infections from the second week up to the end of the first month, and bronchopulmonary dysplasia (BPD) afterwards. Birthweight < 1000 g, gestational age < 30 weeks, absence of spontaneous respiratory activity, unknown body temperature and pH < 7.20 at admission were associated with death from respiratory problems and IVH. Male sex, birthweight < 1000 g and unknown body temperature at admission were associated with death from BPD. Mortality from infections was higher in one centre; no other differences emerged among the eight NICUs. The classification of the causes of death employed and the use of the net probabilities of death appear as practical and useful instruments to study the relationship between specific aspects of medical care and mortality, and to investigate the reasons for differences in performance between neonatal units.