Over a 34-month period, 1105 newborns weighing between 501 and 2000 g at birth were enrolled in a prospective study of the aetiology and consequences of neonatal brain haemorrhage. The three participating hospitals care for approximately 85% of births in the study weight range in Middlesex, Monmouth and Ocean counties, New Jersey. Cranial ultrasonographic imaging through the anterior fontanelle was carried out a mean age of 4.9 +/- 2.2 hours, 25.5 +/- 4.8 hours and 7.2 +/- 0.8 days to detect haemorrhage and other brain lesions. In 93.2% of study infants, scans were read by two independent expert readers (blind to the clinical status of the child) with submission of the scan to a third reader in cases of disagreement. Confirmation of both presence or absence and, when present, scan of first diagnosis of germinal matrix and/or intraventricular haemorrhage (GM/IVH) by two independent readers was achieved in 76.3% of study infants. The first two readers agreed as to presence or absence of GM/IVH in 82.4% of infants (Kappa = 0.56). Interobserver agreement was affected by the reported scan quality and by the number of scans available, but not by the hospital of origin, race or birthweight of the infant.