Understanding the effect of changes in physical activity on mortality risk may help researchers clarify intervention strategies. This study investigated associations of physical activity at baseline and 5 years previously with all-cause mortality risk in a cohort of 472 elderly Dutch men. Relative risks were estimated for relations between mortality in 1990-1995 and physical activity levels in 1990 and 1985. Adjustments were made for baseline age, chronic diseases, functional status, and lifestyle factors. In contrast to previous levels of physical activity (adjusted p-trend = 0.39), baseline total time spent in physical activity was inversely associated with mortality risk (p-trend = 0.004; for the most active tertile vs. the least active, relative risk = 0.44; 95% confidence interval: 0.25, 0.80). No consistent associations were observed after fractionating total physical activity into activities of differing intensity or into four different types of activity. Relative to maintaining a physically active lifestyle (i.e., walking or bicycling for 20 minutes at least three times per week) in both survey years, a gradient of increasing risk was observed from adopting an active lifestyle to becoming sedentary to remaining sedentary (p-trend = 0.004). Recent levels of physical activity were more important for mortality risk among elderly men than activity 5 years previously. Becoming or remaining sedentary was significantly associated with increased mortality risk in comparison with remaining physically active.