We analyzed the association of education, occupation, and sex with incidence of cognitive impairment using data from three communities in the Established Populations for Epidemiologic Studies of the Elderly (EPESE) projects (New Haven, East Boston, and Iowa). Participants were initially interviewed in 1981-1983, with follow-up 3 and 6 years later. Incident cognitive impairment was defined on the basis of either: (1) increase in the number of errors in Short Portable Mental Status Questionnaire (SPMSQ) (i.e. from a baseline level below the cutoff value to a score above the cutoff), or (2) inability to respond to interview questions at a follow-up contact (requiring a proxy informant), or (3) death with a recorded diagnosis of a dementing illness. In multiple logistic regression models, the major factors predicting the development of cognitive impairment were advanced age, any errors on baseline SPMSQ, 8 or fewer years of education, and occupation. Education and occupation remained significant predictors after controlling for age, site, sex, stroke, and baseline SPMSQ score.