In 1968-1969, a population-based sample of Swedish women aged 38-60 years was recruited for a health survey, and 20-year survival was later ascertained from national registries. Occupational and leisure-time physical activity data from the baseline and 6-year follow-up examinations were evaluated in relation to all-cause mortality among 1,405 women who were initially free of major diseases. In comparison with being inactive, the mortality relative risk associated with being somewhat active was 0.28 (95% confidence interval 0.17-0.46) for occupational activity and 0.56 (95% confidence interval 0.39-0.82) for leisure-time activity. Being in the most active occupational or leisure activity category further decreased mortality risk to a minor extent. A within-subject decrease in leisure activity over 6 years was also a significant risk factor for all-cause mortality (relative risk = 2.07, relative to no change), although there was no evidence of a benefit from increasing physical activity levels. Since exclusion of early endpoints did not affect the associations in any significant way, underlying illness is unlikely to have played a major role in these analyses. It is concluded that decreases in physical activity as well as low initial levels are strong risk factors for mortality in women, and that their predictive value persists for many years.