The ability to predict long-term outcome on the basis of objective measures made shortly after the birth of an infant has introduced a completely new approach to the investigation of the aetiology of childhood impairments and disabilities, and the evaluation of the effects of perinatal management regimes designed to avoid or ameliorate them. This approach also has important implications for the management of sick and vulnerable infants, both in the perinatal period and later. From the use of ultrasound brain scanning, a great deal has been learnt about all aspects of the aetiology, evolution and prognosis of GLH, IVH and IPH in very preterm infants; and from autopsy correlation studies, about the underlying pathological processes causing the lesions. There are, however, three observations which have particular implications for the choice of neonatal measure to be used for predictive purposes and for the interpretation of results. First, the long-term prognosis for infants with uncomplicated GLH and IVH who never develop ventricular dilatation, hydrocephalus or evidence of loss of brain tissue is indistinguishable from that of very preterm infants of similar gestational age with no detectable lesions on the ultrasound scan. Second, hypoxic-ischaemic injury is a much more common cause of the neurodevelopmental impairments seen at follow-up in sick or preterm infants than haemorrhage. Third, ultrasound is very poor at detecting hypoxic-ischaemic lesions in the absence of bleeding. These lesions are only reliably recognized with ultrasound when the necrotic process is sufficiently advanced for loss of brain tissue or cerebral atrophy to occur. As a consequence, hypoxic-ischaemic injury is only likely to be identified in the early stages of its evolution following the causal insult by demonstrating deranged cerebral intracellular energy metabolism or cerebral haemodynamics. From these observations it may be deduced that ultrasound brain scanning in the first week of life is a poor predictor of adverse neurodevelopmental outcome at follow-up, depending largely on the diagnosis of haemorrhagic lesions for its 'power'. According to the size of haemorrhage and the outcome considered, the sensitivity may be calculated at 61-73.5% and the positive predictive value at only 32-50% (Stewart ef al, 1987; Cooke, 1987). Nevertheless, the marked echodensities which indicate intraparenchymal haemorrhage (IPH or haemorrhagic PVL) carry a very bad prognosis and thus detect a small group of infants with a probability estimate of serious neurodevelopmental impairment of about 90% (Table 1).(ABSTRACT TRUNCATED AT 400 WORDS)