BACKGROUND: Prospective studies mostly with European and North-American populations have shown inconsistent results regarding the association of overweight/obesity and mortality in older adults. Our aim was to investigate the relationship between overweight/ obesity and mortality in an elderly Brazilian population. METHODS AND FINDINGS: Participants were 1,450 (90.2% from total) individuals aged 60 years and over from the community-based Bambui (Brazil) Cohort Study of Ageing. From 1997 to 2007, 521 participants died and 89 were lost, leading to 12,905 person-years of observation. Body mass index (BMI) and waist circumference (WC) were assessed at baseline and at the 3rd and 5th years of follow-up. Multiple imputation was performed to deal with missing values. Hazard ratios (HR) of mortality for BMI or WC alone (continuous and categorical), and BMI and WC together (continuous) were estimated by extended Cox regression models, which were fitted for clinical, socioeconomic and behavioral confounders. Adjusted absolute rates of death at 10-year follow-up were estimated for the participants with complete data at baseline. Continuous BMI (HR 0.85; 95% CI 0.80-0.90) was inversely related to mortality, even after exclusion of smokers (HR 0.85; 0.80-0.90), and participants who had weight variation and died within the first 5 years of follow-up (HR 0.83; CI 95% 0.73-0.94). Overweight (BMI 25-30 kg/m(2)) was inversely (HR 0.76; 95%CI 0.61-0.93) and obesity (BMI >/= 30 kg/m(2); HR 0.85; 95% CI 0.64-1.14) not significantly associated with mortality. Subjects with BMI between 25-35 kg/m(2) (23.8-25.9%) had the lowest absolute rates of death at 10-years follow-up. The association between WC and death was not significant, except after adjusting WC for BMI levels, when the relationship turned into marginally positive (HR 1.01; CI 95% 1.00-1.02). CONCLUSIONS: The usual BMI and WC cut-off points should not be used to guide public health and clinical weight control interventions in elderly in Brazil