Hurricanes and tropical storms served as natural experiments for investigating whether autism is associated with exposure to stressful events during sensitive periods of gestation. Weather service data identified severe storms in Louisiana from 1980 to 1995 and parishes hit by storm centers during this period. Autism prevalences in different cohorts were calculated using anonymous data on birth dates and parishes of children diagnosed with autism in the state mental health system, together with corresponding census data on all live births in Louisiana. Prevalence increased in dose-response fashion with severity of prenatal storm exposure, especially for cohorts exposed near the middle or end of gestation (p < 0.001). Results complement other evidence that factors disrupting development during sensitive gestational periods may contribute to autism.