STUDY OBJECTIVE: To evaluate the relation of body mass index (BMI) to short-term mortality in a large Italian population sample. DESIGN: Within the Italian RIFLE pooling project, BMI was measured in 47 population samples made of 32,741 men and 30,305 women ages 20-69 years (young 20-44, mature 45-69). Data on mortality were collected for the next six years. MAIN OUTCOME MEASURES: Age adjusted death rates in quintile classes of BMI and Cox proportional hazards models with six year all causes mortality as end point, BMI as covariate and age, smoking, systolic blood pressure as possible confounders were computed. Multivariate analysis was tested in all subjects and after the exclusion of smokers, early (first two years) deaths, and both categories. RESULTS: The univariate analysis failed to demonstrate in all cases a U or inverse J shaped relation. The Cox coefficients for the linear and quadratic terms of BMI proved significant for both young and mature women. The minimum of the curve was located at 27.0 (24.0, 30.0, 95% confidence limits, CL) and 31.8 (25.5, 38.2, 95% CL) units of BMI, for young and mature women respectively. Similar findings were obtained even when exclusion were performed. No relation was found for young men while for mature adult men only the model for all subjects retained significant curvilinear relation (minimum 29.3; 22.4, 36.2, 95% CL). CONCLUSION: These uncommon high values of BMI carrying the minimum risk of death seems to be in contrast with weight guidelines. A confirmation of these findings in other population groups might induce the consideration of changes in the suggested healthy values of BMI.