Objective To explore the association between body mass index (BMI) and all-cause mortality among the elderly in Beijing. Methods This analysis was based on the Beijing multidimensional longitudinal study of aging (BLSA), which included 2,090 subjects over 55 years old and was followed-up from 1992 to 2012. BMI-mortality curves were drawn to find the optimal BMI range with the lowest mortality. Cox proportional hazard models were used to obtain the hazard ratios (HRs) for BMI and BMI changes in the overall population and in specific stratified populations. Results During follow-up, 1,164 deaths were recorded; BMI-mortality curve was U-shaped, with the lowest mortality at a BMI of approximately 25 kg/m(2). After adjusting for gender, age, smoking, drinking and some pre-existing diseases, HRs for underweight, overweight and obesity compared with normal weight were 1.372 (95% CI: 1.154-1.631), 0.767 (95% CI: 0.666-0.884) and 0.871 (95% CI: 0.830-1.246), respectively. HR for BMI drop was 3.245 (95% CI: 0.824-12.772) in the underweight group and 1.892 (95% CI: 0.830-1.246) in the normal weight group, HR for BMI rise was 1.795 (95% CI: 1.243-2.591) in normal weight group and 1.962 (95% CI: 1.202-3.203) in the overweight group. Conclusion Keeping BMI in an overweight status and stable is related to a reduced mortality.