Paquid is an epidemiological study designed to gather and follow up a cohort of 4,000 elderly subjects (65 years and older) living at home in order to study normal and pathological brain aging. These subjects were randomly chosen in the general population of 75 communities of South-Western France. We present the results of the data collected from 2,792 subjects on the prevalence and the correlates of clinically diagnosed dementia. The DSM III criteria for dementia were met by 101 subjects (3.62 p. 100). These cases were reviewed by a neurologist to confirm the diagnosis and to determine the cause of dementia using the NINCDS-ADRDA criteria. Forty-three subjects were classified as probable Alzheimer's disease; 8 as possible Alzheimer's disease; 5 as vascular dementia; 5 as Parkinson's disease with dementia; 2 as alcoholic dementia; 2 as "dementified psychosis"; and 1 unclassified. Fifteen patients refused to be examined by the neurologist, 18 were false-positives, and 2 died before the neurologists visit. Using the NINCDS-ADRDA criteria, the prevalence of dementia was as low as 1.6 p. 100. The prevalence of probable Alzheimer's disease decreased dramatically as educational level increased, lung 5.4 p. 100 for subjects with no education, 1.7 p. 100 for subjects with grade school level, 0.4 p. 100 for subjects with high school level and 0.4 p. 100 for subjects with university degrees. The relationship between dementia and educational level is still controversial in the literature. In this study the sample was large and randomly selected; all the demented cases fulfilled the NINCDS-ADRDA criteria. This suggests that educational level is indeed an important correlate of dementia in the French elderly community.