Children aged 0–14 on a specified census day, with impairments of verbal and nonverbal communication, social interaction and imaginative play, and with repetitive, stereotyped activities, were identified in an epidemiological study carried our in an area of southeast London. Sociable moderately, severely, and profoundly retarded children were included for comparison. An overall male:female ratio of 2:6:1 was found in those with language and social impairments, but, in the children of this group who were moderately, severely, or profoundly retarded, the ratio was 2:1:1, closely similar to that found in children in the same IQ range with Down's syndrome or cerebral palsy. The excess of males was much more marked in language and socially impaired children who were of higher ability, or who had a history of typical early childhood autism. The findings were linked to hypotheses of genetically greater variability in males, and to male-female differences in visuo-spatial and language skills.