Changes in cognitive function were investigated in 566 subjects 65-86 years old at baseline, who are a subsample of the Western Collaborative Group Study, a cardiovascular epidemiologic study of middle-aged men that began in the 1960s. Cognitive function was assessed in 1986-1988 (baseline) and again in 1992-1994 by three standardized measures: the Benton Visual Retention Test, the Controlled Oral Word Association Test, and the Digit Symbol Substitution (DSS) Test. Longitudinal change in performance was defined as the shift over time in a subject's quartile rank ordering, using the baseline distribution of test scores as a standard. 'Decliners' and 'improvers' in cognitive function were subjects who lost or gained, respectively, two or more quartile ranks on all three tests combined. By this definition, 20% (n = 113) of subjects declined, compared with 17% (n = 95) who improved in cognitive performance from 1986-1988 to 1992-1994. After adjustment for age, education, and physical health, decline in cognitive performance was significantly associated with poor self-perceived health ratings, depression scale scores, and self-reports of physical activity. Rank score change in the DSS Test was the single best predictor of cognitive function at follow-up on a diverse battery of neuropsychological tests.