Using the unique resources of the Rochester Epidemiology Project for population-based studies, we identified 629 Olmsted County, Minn., residents who fulfilled the 1988 International Headache Society criteria for newly diagnosed migraine over a 3-year period. Over 6,400 patient records from several diagnostic rubrics were screened; a substantial proportion of cases had been 'signed-out' to diagnoses other than 'migraine headache'. Medical records were reviewed by two trained nurses who abstracted supporting data for two neurologists. The neurologists determined whether each case met eligibility requirements and assigned a headache diagnosis by consensus. The diagnostic criteria offered some flexibility and were adapted to retrospective record-based research. Most records contained enough information to effectively classify the headache, although information on the frequency and duration of attacks proved to be problematic. A validation re-abstraction of a 10% sample of cases was undertaken with acceptable reproducibility of symptoms and diagnosis. Our study shows that migraine headache can be studied retrospectively through existing detailed medical records.