There is considerable controversy over reasons for observed increases in the prevalence of autism spectrum disorders. We examined trends in British Columbia education database coding of children with autism from 1996 to 2004. There was a significant linear increase in autism prevalence. The proportion of children identified by age 6 increased significantly from 1996 to 1999. When we calculated prevalence assuming onset prior to age 3, previously unidentified cases, particularly among girls in 1996 and 1997, accounted for substantial increases in estimated prevalence. The magnitude of under-identification decreased from 1996 to 2000, and rose slightly in 2001. Analyses of prevalence trends must take into account effects of earlier age at identification and inclusion of previously undetected cases on prevalence estimates.