Most studies on the frequency of autism have had methodological problems. Most notable of these have been differences in diagnostic criteria between studies, degree of cases overlooked by the initial screening, and type of measurement. This study aimed to replicate the first report on childhood autism to address cumulative incidence as well as prevalence, as defined in the International Statistical Classification of Diseases and Related Health Problems, 10th revision (ICD-10) Diagnostic Criteria for Research. Here, the same methodological accuracy (exactness of a measurement to the true value) as the first study was used, but population size was four times larger to achieve greater precision (reduction of random error). A community-oriented system of early detection and early intervention for developmental disorders was established in the northern part of Yokohama, Japan. The city's routine health checkup for 18-month-old children served as the initial mass screening, and all facilities that provided child care services aimed to detect all cases of childhood autism and refer them to the Yokohama Rehabilitation Center. Cumulative incidence up to age 5 years was calculated for childhood autism among a birth cohort from four successive years (1988 to 1991). Cumulative incidence of childhood autism was 27.2 per 10 000. Cumulative incidences by sex were 38.4 per 10 000 in males, and 15.5 per 10 000 in females. The male: female ratio was 2.5:1. The proportions of children with high-functioning autism who had Binet IQs of 70 and over and those with Binet IQs of 85 and over were 25.3% and 13.7% respectively. Data on cumulative incidence of childhood autism derived from this study are the first to be drawn from an accurate, as well as precise, screening methodology.