This paper describes the method of compilation of the Northern Ireland Cerebral Palsy Register (NICPR) and outlines the epidemiology of cerebral palsy (CP) in Northern Ireland in 1981--93 based on an analysis of 784 cases. Multiple and overlapping sources of notification were used to compile the NICPR. Each case was followed up with a standardised assessment completed at age 5 years and validated by a paediatrician. Over 70% of cases were multiply notified, and paediatricians were the most important source of notification. Cases of CP most likely to be under-ascertained included those very mildly affected, cases who died before the register began (death certificates were searched but found to be unreliable) and cases who moved out of the area before their diagnosis was confirmed. The prevalence of CP for the birth period 1981--93 was 2.24 per 1000 livebirths [95% CI 2.08, 2.40]. There were no statistically significant temporal or geographic variations in the rate of CP. Almost half the cases of CP were of low birthweight (<2500 g), and this proportion increased during 1987--89, but decreased (P < 0.01) more recently (1990--93). The proportion of cases weighing <1000 g trebled from 3% in 1981--83 to 10% in 1990--93 (P < 0.01). The most common CP subtype was bilateral spastic cerebral palsy (55%). Just over a quarter (29%) of cases were unable to walk (with/without aids), and one-fifth (22%) had no useful hand/arm function. Almost half (49%) the cases had at least one other impairment (intellectual, sensory impairment or active seizures) in association with their CP. Prevalence rates and the pattern of disability in populations of people with CP can only be evaluated on the basis of the methods of case definition and ascertainment. We have demonstrated the methods necessary to ensure valid, standard and detailed information on CP for the purposes of surveillance, service planning and research.