The objective of this study was to compare two methods of measuring physical function in subjects with a broad range of abilities and to evaluate the effects of cognitive, social, educational, and age factors on the relationship between the two methods. Multiple regression analysis was used to compare self-perceived (dependent variables) with performance measures (independent variables). Covariates included age, gender, Mini-Mental State Exam score, education, living status, and depression score. Five community-dwelling and two nursing home sites. 417 community-dwelling subjects and 200 nursing home residents aged 62-98 years. Self-perceived physical function was assessed with the physical dimension summary score of the Sickness Impact Profile, which comprises three subscales: ambulation, mobility, and body care and movement. Physical performance was evaluated by self-selected gait speed, chair-stand time, maximal grip strength, and a balance score. Nursing home residents and community-dwellers were significantly different (P < .0001) in all variables except age and gender. Self-perceived and performance-based measures were moderately correlated, with a range from r = -.194 to r = -.625 (P < .05). Gait speed was the strongest independent predictor of self-perceived physical function in both groups. Symptoms of depression were also an independent predictor of self-perceived function in nursing home residents; subjects who had such symptoms report more self-perceived dysfunction than would be predicted based on performance tests. Self-selected gait speed is a global indicator of self-perceived physical function over a broad range of abilities. External determinants (depressive symptoms, cognitive function, marital status, etc.) affect self-perceived function in both groups, but gait speed is the greatest single predictor of self-perceived function. In nursing home residents depressive symptomatology is related to self-perceived.