The prevalence of autism spectrum disorders (ASD) from the 2007 National Survey of Children's Health (NSCH) was twice the 2003 NSCH estimate for autism. From each NSCH, we selected children born in the US from 1990 to 2000. We estimated autism prevalence within each 1-year birth cohort to hold genetic and non-genetic prenatal factors constant. Prevalence differences across surveys thus reflect survey measurement changes and/or external identification effects. In 2003, parents were asked whether their child was ever diagnosed with autism. In 2007, parents were asked whether their child was ever diagnosed with an ASD and whether s/he currently had an ASD. For the 1997-2000 birth cohorts (children aged 3-6 years in 2003 and 7-10 years in 2007), relative increases between 2003 autism estimates and 2007 ASD estimates were 200-600 %. For the 1990-1996 birth cohorts (children aged 7-13 years in 2003) increases were lower; nonetheless, differences between 2003 estimates and 2007 "ever ASD" estimates were >100 % for 6 cohorts and differences between 2003 estimates and 2007 "current ASD" estimates were >80 % for 3 cohorts. The magnitude of most birth cohort-specific differences suggests continuing diagnosis of children in the community played a sizable role in the 2003-2007 ASD prevalence increase. While some increase was expected for 1997-2000 cohorts, because some children have later diagnoses coinciding with school entry, increases were also observed for children ages ≥ 7 years in 2003. Given past ASD subtype studies, the 2003 "autism" question might have missed a modest amount (≤ 33 %) of ASDs other than autistic disorder.