Studies of the role of leptin in patients with anorexia nervosa and bulimia nervosa have conflicted in their data and interpretation. Such differences may be a result of the assay methods used or the way results are compared with those from normal controls. To investigate these possibilities, we analyzed serum leptin levels in anorexic, bulimic, obese, and control individuals, thereby spanning the full range of human body weights, using three frequently employed commercial kits. Kits from Linco (St Louis, MO) and DSL (Webster, TX) employ a radioimmunoassay method, and the R&D Systems kit (Minneapolis, MN) uses an enzyme-linked immunosorbent assay. We found that the three kits provide results that are highly linearly correlated with each other and remarkably linearly related to percent ideal body weight (%IBW) over more than three orders of magnitude (Linco, r = 0.90; R&D, r = 0.87; DSL, r= 0.86). For very low leptin levels, the more sensitive kits from R&D and Linco appeared to give more reliable results. Measurement method does not appear to explain the literature conflicts. We found that patients with anorexia nervosa have serum leptin values that lie above the line extrapolated from the %IBW/leptin curve generated from analysis of all nonanorexic patients. Therefore, in anorexia nervosa, inappropriately high leptin levels for %IBW may contribute to a blunted physiologic response to underweight and consequent resistance to dietary treatment. By contrast, most bulimic patients have leptin levels significantly below those predicted from the same %IBW/leptin curve. The relative leptin deficiency in bulimic subjects may contribute to food-craving behavior. We propose that using the %IBW/leptin curve can facilitate identification of true pathophysiologic abnormalities in eating-disordered individuals and provide a basis for the design of therapeutic interventions or monitoring of response to treatment.