Body composition changes as a result of ageing may impact the survival of older adults. However, its influence on mortality risk is uncertain. Currently, the best method for body composition analysis in clinical practice is DXA. Nonetheless, the few studies on body composition by DXA and mortality risk in the elderly have some limitations. We investigated the association between body composition by DXA and mortality in a cohort of elderly subjects. Eight hundred thirty‐nine community‐dwelling subjects (516 women, 323 men) ≥ 65 years of age were assessed by a questionnaire, clinical data, laboratory exams, and body composition by DXA at baseline. Total fat and its components (eg, visceral adipose tissue [VAT]) were estimated. Appendicular lean mass (ALM) adjusted for fat and ALM divided by height² was used to ascertain the presence of low muscle mass (LMM). Mortality was recorded during follow‐up. Multivariate logistic regression was used to compute ORs for all‐cause and cardiovascular mortality. Over a mean follow‐up of 4.06 ± 1.07 years, there were 132 (15.7%) deaths. In men, after adjustment for relevant variables, the presence of LMM (OR, 11.36, 95% CI, 2.21 to 58.37, P = 0.004) and VAT (OR, 1.99, 95% CI, 1.38 to 2.87, P < 0.001, for each 100‐g increase) significantly increased all‐cause mortality risk, whereas total fat, measured by the fat mass index, was associated with decreased mortality risk (OR, 0.48, 95% CI, 0.33 to 0.71, P < 0.001). Similar results were observed for cardiovascular mortality. In women, only LMM was a predictor of all‐cause (OR, 62.88, 95% CI, 22.59 to 175.0, P < 0.001) and cardiovascular death (OR, 74.54, 95% CI, 9.72 to 571.46, P < 0.001). LMM ascertained by ALM adjusted for fat and fat mass by itself are associated with all‐cause and cardiovascular mortality risk in the elderly. Visceral and subcutaneous fat have opposite roles on mortality risk in elderly men. Thus, DXA is a promising tool to estimate risk of mortality in the elderly. © 2019 American Society for Bone and Mineral Research.