Sixteen studies of the prevalence of autism in childhood, using epidemiological methods in defined populations in Europe, the USA and Japan, in English or with English summaries, were found in the published literature. Age specific rates varied from 3.3 to 16.0 per 10,000. Eight studies gave rates for a sub-group of 'typical' autism varying from 1.2 to 8.4. Reasons for variations were sought by examining geographic and demographic details of the populations screened, methods for initial screening and final examination of possibly autistic children, demographic and clinical details of children identified as autistic, and criteria used for diagnosis. There was evidence, independent of diagnostic criteria, of a higher prevalence among children of first generation immigrants to Europe from 'exotic' countries. Apart from this, all differences could be due to variations in diagnostic practice and increasing awareness of the manifestations of autistic conditions throughout the range of intelligence, from severely retarded to average and above. However, it remains possible, though not proven, that rates do vary over place and/or time. The problems of defining a sub-group with 'typical' autism among the wide spectrum of children with the triad of impairments of social interaction, communication and imagination are discussed and the value of such a sub-grouping questioned.