BACKGROUND: Study of the influence of physical activity and body composition in the context of the disablement process requires analytic techniques that can address time-dependent confounding related to exposures for the occurrence of functional limitation and disability. METHODS: We applied logistic marginal structural models to explore causal relationships between leisure time physical activity and a measure of relative muscle mass (the lean:fat mass ratio) and self-reported functional limitation. Subjects (n = 1655) were members of a population-based cohort recruited in Sonoma, California, in 1993-1994 (median age 70 years); they were surveyed 3 times over 6.5 years. RESULTS: Based on the marginal structural model, the causal odds ratio for functional limitation with a 0.5-unit increment in relative muscle mass was 0.56 in women (95% CI = 0.46-0.67). This reduction in odds was not altered by adjustment for baseline covariates. The corresponding causal odds ratio in men was 0.77 (0.65-0.92). This also was not influenced by baseline covariates or levels of physical activity. There was evidence for an independent causal effect of increased levels of physical activity on reduction of odds of functional limitation for men but not for women. Obese women experienced no protective effect of increased lean-to-fat mass. Normal-weight women experienced a benefit with increasing levels of physical activity. CONCLUSIONS: Marginal structural models provide a means to address time-dependent confounding, which can occur in longitudinal studies. These analyses indicate that leisure time physical activity exerts its beneficial effects through reductions in fat mass relative to lean body mass.