To evaluate the correlation between body mass index (BMI), body composition, and all-cause mortality in an elderly Asian population. A prospective observational cohort study with 3.5-year follow-up. The Korean Longitudinal Study on Health and Aging Project for elderly residents in Seongnam City, Korea. Eight hundred seventy-seven subjects aged 65 and older for whom baseline body composition data was available. BMI, waist circumference, and body composition of each subject was evaluated. Body composition was examined using bioelectrical impedance analyses of measures, including lean mass (kg), fat mass (kg), and fat proportion (%). In addition, lean mass index (LMI, kg/m(2)) was calculated by dividing lean mass by the square of height. Participants were divided into three groups: Group 1 (<25th percentile), Group 2 (25-75th percentiles), and Group 3 (> or =75th percentile) for BMI, waist circumference, body composition, and LMI. In the fully adjusted Cox proportional hazard model, BMI, waist circumference, and fat composition were not correlated with mortality, but higher lean mass and LMI were considered predictors of lower mortality when comparing Group 3 and Group 1 (in lean mass, relative risk reduction of 84%, 95% confidence interval (CI)=45-96%, P=.004; in LMI, relative risk reduction of 69%, 95% CI=12-89%, P=.03). The present study indicates that the recommendation of low BMI as a means of obtaining a survival advantage in the elderly is not supported. Instead, higher lean mass and higher LMI are associated with better survival in the elderly Asian population.