We used the resources of the Rochester Epidemiology Project to compare sociodemographic and clinical characteristics in three groups of Alzheimer's disease patients. The first group included incidence cases occurring among residents of Rochester, Minnesota (population-based series; n = 241). The second group was a sample of patients referred to the Mayo Clinic form the remainder of Minnesota and the four surrounding states (n = 58); the third was a sample referred from the remainder of the United States (n = 94). Patients from Rochester were more frequently women, less highly educated, less commonly white collar workers, more frequently institutionalized, less frequently married, and more often lived along than those in the two referral groups. Patients from Rochester also had a more advanced age of onset of dementia. For occupation, education, and living arrangement, the differences across groups increased with increasing distance of referral. Clinical and epidemiological studies based on patients referred form primary to secondary or tertiary care centers may suffer from severe selection bias.