OBJECTIVES:  To estimate treatment costs attributable to overweight or obese status in diabetic patients in the U.S.A. METHODS:  The data was drawn from the 2003-2006 Medical Expenditure Panel Survey.  Adult patients (18-74 years-old) with diabetes were identified based on a self-reported diagnosis or ICD-9-CM code of 250.  Patients with pregnancy, malignancy, kidney dialysis, immunodeficiency, or body-mass-index (BMI) <18.5 were excluded.  Medical treatment costs included office based physician/outpatient visits, emergency room visits, or hospitalizations, excluding dental problems and injuries. The treatment costs attributable to being overweight (25≤BMI<30) or obese (BMI≥30) at various points of the cost distribution were estimated using weighted quantile regression after controlling for demographics, comorbidities, and other study variables.  Treatment costs attributable to being overweight/obese were calculated by the differences in the actual treatment costs for overweight/obese patients and the expected costs if obese patients were normal-weight patients using the study variable coefficients obtained from all patients.  All costs were converted to 2006 U.S. dollars using price indices.  Data were analyzed using SAS and SUDAAN. RESULTS: A total of 5338 patients with diabetes were selected for this study.  Approximately 88% of the diabetic patients were overweight (31%) or obese (57%).  Compared with normal-weight patients, the incremental treatment costs attributable to obesity were significantly higher by $35, $96, $196, $290, and $739 at the 10th, 25th, 50th, 75th, and 90th percentiles, respectively.  Similar trends were found in overweight patients compared with those of normal-weight, but the attributable costs were not significantly different except at the 75th percentile point. CONCLUSIONS: The overall economic burdens attributable to being overweight and obese in diabetic patients were substantial and increased significantly in the upper tail of the treatment cost distribution.  The important finding that the magnitude of attributable costs increased across the distribution of treatment costs would not have been possible without the quantile regression method.