A brief diagnostic battery of neuropsychological tests was developed for a large-scale epidemiological study of dementia. We operationally defined dementia as defective memory and defective performance in at least two other areas, including orientation, abstract reasoning, construction, and language. Criterion scores for defining defective performance on each test were developed. In a pilot study that used 51 different subjects with a working diagnosis based on physicians' assessment (ie, 32 demented and 29 nondemented subjects), the test-based diagnosis agreed with the working diagnosis in all but two cases. The test battery was then applied to 430 healthy elderly subjects. Eighteen percent of those with 8 or less years of education met criteria for dementia compared with 5% of those with more than 8 years of education. We computed education-corrected scores for each test with the use of residuals from the regression of each test score on education. Based on corrected scores, 12 subjects were reclassified as nondemented and 11 as demented. Subjects who were reclassified as demented were significantly more impaired in activities of daily living than nondemented subjects who were not reclassified. Activities of daily living in subjects who were reclassified as nondemented did not differ from those in demented subjects who were not reclassified. These findings suggest that the neuropsychological battery may have utility in the diagnosis of dementia. However, neuropsychological performance may be influenced by education, and some form of adjustment, such as correction for activities of daily living, may be required in epidemiological studies.