This study used a linear structural relations modeling technique (LISREL) to examine longitudinal data for 1,192 persons from a community-based population. The goal was to test the ability of an a priori model to predict cognitive change over a 2.0- to 2.5-year period in older adults aged 70-79 at the initial evaluation. The model included 22 demographic, physical, and psychosocial variables as predictors of cognitive function and cognitive change. The study used an exploratory-confirmatory design, enabling cross-validation of the model developed in the exploratory set in the confirmatory sample. Structural equation modeling analyses identified 4 endogenous model variable (education, strenuous activity, peak pulmonary expiratory flow rate, and self-efficacy) as direct predictors of cognitive change over the study period.