Background: The hypothesis of reverse epidemiology holds that, obesity may reduce the risk of clinical adverse events in older subjects. However, this association is controversial and rarely explored according to the underlying health status. We tested this phenomenon by assessing the association between body mass index (BMI) and clinical adverse events in community dwelling older women according to their frailty status. Methods: EPIDOS is a multicenter prospective cohort of community-dwelling women aged 75 and older recruited between 1992 and 1994. At baseline, we collected demographics, BMI (<21 kg/m2: underweight; 21-24.9: normal weight; 25-29.9: overweight and ≥30: obesity), frailty through Fried model, and clinical characteristics. All-cause mortality, falls, hip fractures, and hospital admission were collected within 5 years of follow-up and were analyzed using univariate and multivariate survival analysis by using Kaplan-Meier methods and Cox Hazard Proportional models. Results: Of 6662 women (mean age, 80.4 years), 11.6%; 95% Confidence Interval (95% CI) CI [10.8%-12.3%] were frail. By multivariate analysis, the risk of death in frail women (compared to not-frail normal weight women) decreases with increase of BMI: adjusted Hazard Ratio (aHR)frail-underweight = 2.04 [1.23-3.39]; aHRfrail-normal weight = 3.07 [2.21-4.26]; aHRfrail-overweight = 1.83 [1.31-2.56]; aHRfrail-obese = 1.76 [1.15-2.70]; p < 0.001. Frail overweight and obese women had a significant lower risk of death than frail normal-weight women (p = 0.004). Similar features were found for fall risk and hip fracture and for not-frail women. The relative risks of hospital admission for normal weight, overweight and obese frail women were similar (aHRfrail-normal weight = 1.50 [1.22-1.84], aHR frail-overweight =1.48 [1.26-1.74] and aHR frail-obese =1.53 [1.24-1.89], respectively). Conclusion: Our results suggest that overweight and obesity reduce the risks of clinical adverse events in frail community-dwelling older women and that frailty definition through Fried model had to be re-calibrated for overweight and obese individuals.