OBJECTIVES: Previous studies of weight change and mortality in older adults have relied on self-reported weight loss, have not evaluated weight gain, or have had limited information on health status. Our objective was to determine whether 5% weight gain or loss in 3 years was predictive of mortality in a large sample of older adults. DESIGN: Longitudinal observational cohort study. SETTING: Four U.S. communities. PARTICIPANTS: Four thousand seven hundred fourteen community-dwelling older adults, age 65 and older. MEASUREMENTS: Weight gain or loss of 5% in a 3-year period was examined in relationship to baseline health status and interim health events. Risk for subsequent mortality was estimated in those with weight loss or weight gain compared with the group whose weight was stable. RESULTS: Weight changes occurred in 34.6% of women and 27.3% of men, with weight loss being more frequent than gain. Weight loss was associated with older age, black race, higher weight, lower waist circumference, current smoking, stroke, any hospitalization, death of a spouse, activities of daily living disability, lower grip strength, and slower gait speed. Weight loss but not weight gain of 5% or more was associated with an increased risk of mortality that persisted after multivariate adjustment (Hazard ratio (HR) = 1.67, 95% CI = 1.292.15) and was similar in those with no serious illness in the period of weight change. Those with weight loss and low baseline weight had the highest crude mortality rate, although the HR for weight loss was similar for all tertiles of baseline weight and for those with or without a special diet, compared with those whose weight was stable. CONCLUSIONS: This study confirms that even modest decline in body weight is an important and independent marker of risk of mortality in older adults.