This population-based, retrospective cohort study of neonatal seizures included all neonates born to residents of Fayette County, Kentucky, from 1985 to 1989. We ascertained potential cases by computer search of hospital-based medical record systems, Kentucky Center for Health Statistics birth certificate data files, and National Center for Health Statistics multiple-cause-of-death mortality data files. Medical records for potential cases were abstracted, and relevant portions were reviewed independently by three neurologists using prospectively determined case-selection criteria. Seizures occurred in 58 of 16,428 neonates (3.5/1,000 live births). An additional 15 neonates had possible seizures, for a combined risk of 4.4/1,000 live births. Neonatal seizure risk varied inversely with birth weight: 57.5/1,000 live births among very low birth weight infants (< 1,500 grams) compared with 4.4/1,000 for infants with moderately low birth weight (1,500 to 2,499 grams), 2.8/1,000 for those with normal birth weight (2,500 to 3,999 grams), and 2.0/1,000 for those with high birth weight (4,000 or more grams). Risk varied among the four hospitals in the county with obstetric units, the university hospital having the highest risk. Risk did not differ by race or gender. A Cox proportional hazards model confirmed the results of the simpler univariate analyses. Differences in birth weight of the subpopulations served by each hospital accounted for much but not all the differences in hospital-specific risk.