Background. To reconcile "the obesity paradox," we tested if (1) the contribution of anthropometric measures to mortality was nonlinear and (2) the confounding of hip circumference contributed to the obesity paradox recently observed among diabetic patients. Methods. We analyzed data of diabetic patients attending a community-based prospective, "Tehran lipid and glucose study." In the mortality analysis, anthropometric measures-body mass index (BMI), waist, and hip circumference-were assessed using Cox models incorporating cubic spline functions. Results. During 12 990 person-years follow-up, BMI levels below 27 and those above 40 kg·m(-2) were associated with increased mortality. When we added waist circumference to the BMI in the multivariate-adjusted model, the steepness of BMI-mortality association curve slope for values below 27 kg·m(-2) increased, whereas the steepness of BMI-mortality association curve slope for values above this threshold decreased. Further adjusting the model for hip circumference, the steepness of the slopes of the association curve moved towards null on both extremes and no associations between BMI and all-cause mortality remained. Conclusion. BMI harbors intermixed positive and negative confounding effects on mortality of waist and hip circumference. Failing to control for the confounding effect of hip circumference may stymie unbiased hazard estimation and render conclusions paradoxical.