Recent studies suggest that the prevalence of autism spectrum disorders may be as high as 60 per 10000, considerably greater than the long-accepted figure of 5 per 10 000 for classic autism. Increased recognition, the broadening of the diagnostic concept and methodological differences across studies may account for most or all of the apparent increase in prevalence, although this cannot be quantified. In addition to the implications for families and services, these conceptual changes will affect the scientific study of autism. At present, case definition is reliant on the behavioural and developmental picture alone. Because the behavioural phenotype of autism and the broader autism spectrum disorders includes individuals with different ultimate aetiologies, even when biological or genetic markers are found they will not be present in all individuals with the phenotype. The fact that autism is not a unitary 'disorder' presents a significant challenge to genetic, biological, neurological and psychological research. Progress has recently been made in the earlier identification of autism both through screening programmes and by increased understanding and enhanced surveillance. This offers an opportunity to better understand the early developmental course of autism and may provide additional clues to the underlying pathology.