To investigate the effect of body weight change on longevity.Cohort analytic study, following men from 1977 through 1988.The study was conducted among Harvard University alumni with mean age of 58 years.Alumni, free of cardiovascular disease and cancer, completed questionnaires on weight, height, cigarette habit, and physical activities in 1962 or 1966 and in 1977 (n = 11,703). We assessed weight change between questionnaires, based on self-reported weights.Mortality from all causes (n = 1441), coronary heart disease (n = 345), and cancer (n = 459), determined from death certificates.Lowest all-cause mortality was among alumni maintaining stable weight (+/- 1 kg). With this category as referent (relative risk = 1.00), relative risks of death associated with losing more than 5 kg, losing between 1 and 5 kg, (more than 1 kg and up to 5 kg) gaining between 1 and 5 kg (more than 1 kg and up to 5 kg), and gaining more than 5 kg were 1.57 (95% confidence interval, 1.34 to 1.84), 1.26 (1.10 to 1.46), 1.06 (0.90 to 1.24), and 1.36 (1.11 to 1.66), respectively. For coronary heart disease mortality, relative risks were 1.75 (1.26 to 2.43), 1.43 (1.05 to 1.93), 1.28 (0.91 to 1.80), and 2.01 (1.36 to 2.97), respectively. Weight change did not predict cancer mortality. Findings were not explained by cigarette habit, physical activity, or body mass index. We observed similar trends for follow-up between 1977 and 1982 and between 1983 and 1988. Those losing or gaining more weight also reported greater total lifetime weight loss, which may indicate weight cycling.Both body weight loss and weight gain are associated with significantly increased mortality from all causes and from coronary heart disease but not from cancer.