Objectives: To evaluate the associations of body mass index (BMI) with all-cause, cardiovascular disease (CVD), and expanded CVD mortality in the elderly. Design: Observational cohort study. Setting: Annual physical examination program for the elderly from 2006 to 2010. Participants: We included 77,541 Taipei residents aged ≥65 years (39,365 men and 38,176 women). Measurements: BMI was categorized as underweight (BMI<18.5), normal weight (18.5≤BMI<25), overweight (25≤BMI<30), grade 1 obesity (30≤BMI<35), or grade 2-3 obesity (BMI≥35). Mortality was ascertained by national death files. Results: Underweight (hazard ratios [HRs] of all-cause, CVD, and expanded CVD mortality: 1.92, 1.74, and 1.77, respectively), grade 2-3 obesity (HRs: 1.59, 2.36, and 2.22, respectively), older age, male sex, smoking, and high fasting blood sugar were significant predictors of mortality. Meanwhile, being married/cohabitating, higher education, alcohol consumption, more regular exercise, and high total cholesterol were inversely associated with mortality. Multivariate stratified subgroup analyses verified smokers (HRs of all-cause, CVD, and expanded CVD mortality: 3.25, 10.71, and 7.86, respectively, for grade 2-3 obesity), the high triglyceride group (HRs: 5.82, 10.99, and 14.22, respectively for underweight), and patients with 3-4 factors related to metabolic syndrome (HRs: 4.86, 12.72, and 11.42, respectively, for underweight) were associated with mortality. Conclusion: The associations of BMI with all-cause, CVD, expanded CVD mortality in the elderly are represented by U-shaped curves, suggesting unilateral promotions or interventions in weight reduction in the elderly may be inappropriate. Heterogeneous effects of grades 1 and 2-3 obesity on mortality were observed and should be treated as different levels of obesity.