Recent studies suggested that education may modify the clinical expression of dementia and Alzheimer's disease through its association with a brain reserve capacity. We studied whether education would be related to degree of cognitive decline in mild dementia. Equations to estimate premorbid cognitive ability were derived from a representative normative sample of 83 community-dwelling African Americans using age, education, and gender as independent variables and Word List Learning (WLL) and Animal Fluency (AF) scores from the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) neuropsychological test battery as dependent variables. These equations were applied to a second sample of 131 African Americans (22 with dementia, 109 healthy) who completed CERAD test batteries as part of an epidemiologic study of dementia in the community. Differences between obtained and estimated premorbid WLL and AF test scores were calculated and then analyzed in a 2 (Education) x 2 (Diagnosis) ANOVA. A significant interaction association between Education and Diagnosis on WLL scores and a borderline significant interaction on AF scores showed that the high-education demented group had a greater cognitive decline from estimated premorbid levels than the low-education demented group. Thus, at comparable levels of clinical dementia severity, greater cognitive decline occurred in highly educated patients than in low-educated patients.