Autism prevalence studies have often relied on administrative prevalence or clinical diagnosis as case-identification strategies. We report the incidence of clinical diagnoses of autism spectrum disorders (ASD), versus research-identified autism among residents of Olmsted County, Minnesota, age ≤21 years, from 1976–1997. The incidence of clinically diagnosed ASD (with 95% CI) was 1.5 per 100,000 (0.0–3.7) in 1980–1983 and 33.1 (22.8–43.3) in 1995–1997, a 22.1-fold increase. In contrast, the incidence of research-identified autism increased from 5.5 (1.4–9.5) per 100,000 to 44.9 (32.9–56.9), an 8.2-fold increase. Only 46.8% of research-identified cases received a clinical diagnosis of ASD. These findings demonstrate the potential for misleading interpretation of results from epidemiologic studies that rely on clinical diagnosis of autism to identify cases.