Background: low fat-free mass has been related to high mortality in patients. This study evaluated the relationship between body composition of healthy elderly subjects and mortality. Methods: in 1999, 203 older subjects underwent measurements of body composition by bioelectrical impedance analysis, Charlson co-morbidity index and estimation of energy expenditure through physical activity by a validated questionnaire. These measurements were repeated in 2002, 2005 and 2008 in all consenting subjects. Mortality data between 1999 and 2010 were retrieved from the local death registers. The relationship between mortality and the last indexes of fat and fat-free masses was analysed by multiple Cox regression models. Results: women's and men's data at last follow-up were: age 81.1 ± 5.9 and 80.9 ± 5.8 years, body mass index 25.3 ± 4.6 and 26.1 ± 3.4 kg/m2, fat-free mass index 16.4 ± 1.8 and 19.3 ± 1.9 kg/m2 and fat mass index 9.0 ± 3.2 and 6.8 ± 2.0 kg/m2. Fifty-eight subjects died between 1999 and 2010. The fat-free mass index (hazard ratio 0.77; 95% confidence interval 0.63–0.95) but not the fat mass index, predicted mortality in addition to sex and Charlson index. The multiple Cox regression model explained 31% of the variance of mortality. Conclusion: a low fat-free mass index is an independent risk factor of mortality in elderly subjects, healthy at the time of body composition measurement.