OBJECTIVES: This study tests a flexible modeling approach, which endogenously estimates the non-linear and asymmetric functional form for body mass index (BMI), to examine the relationship between mortality and obesity measured as BMI >30. METHODS: This study used the National Health Interview Survey (NHIS), between 1997 and 2000. Respondents were linked to the National Death Index with mortality follow-up through 2005. We estimated the 5-year probability of death using the logistic regression model adjusting for BMI, age and sex. The multivariable fractional polynomials (MFP) procedure was employed to determine the best fitting functional form for BMI and compared to alternative functional forms using a chi-squared test. Expected years of life lost due to obesity were based on adjusted death probabilites and computed using standard life table functions. RESULTS: The best fitting adjustment model contains the powers -1 and -2 for BMI. A chi-squared test shows a statistically significant improvement in model fit compared to other BMI polynomial functions. The estimated relationship between 5-year probability of death and BMI exhibits a J-shaped pattern for women and a U-shaped pattern for men. The BMI associated with minimum mortality is 27.53 for males and 25.19 for females. A 40-year-old female with a BMI of 40 has an estimated 5.82 fewer years of expected life compared to an analogous female with a BMI of 25. For a comparable change in BMI in a 40-year-old male, the expected years of life lost is 5.20. CONCLUSIONS:The BMI-mortality relation is flat around the minimum, but especially high mortality is associated with the morbidly obese. The MFP approach provides a robust alternative to estimating mortality by allowing the data to determine the best fitting model. The approach is also useful in estimating the relationship between the full spectrum of BMI values and other health outcomes.