OBJECTIVES: To estimate the impact of overweight or obesity on treatment costs(TCs) in diabetic patients in the United States. METHODS: Five-year (2001-2005) pooled Medical Expenditure Panel Survey data, a nationally representative sample of U.S. non-institutionalized population, was used.  A total of 7396 adult diabetic patients were identified based on ICD-9-CM code of 250 after excluding pregnant women and cancer patients.  These patients were classified as normal(body mass index(BMI):18.5-<25), overweight(BMI:25-<30), or obese(BMI:≥30).  TCs included all costs except for treatment of dental problems or injuries.  The impact of overweight/obesity on TCs at various points of the cost distribution was estimated using the weighted quantile regression model after adjusting for age, gender, and other study variables.  The effects of the study variables on the median TCs were investigated using the weighted median regression.  All costs were converted to 2005 U.S. dollars using price indices.  Data were analyzed using SAS and SUDAAN. RESULTS:  Compared with normal-weight patients, the incremental TC attributable to overweight were significantly higher from $238, $268, $409, and $442 at the 10th, 25th, 50th, and 75th percentile respectively.  But incremental costs were diminished to $270 at the 90th percentile because of high costs in normal weight patients with severe comorbidities such as nephropathy.  Similar trends were found in obese-patients compared with normal-patients, and attributable costs are bigger.  Median TCs were increased in women vs. men and Caucasian vs. African-American, and as patients became older.  CONCLUSIONS:  The impact of overweight or obesity on TCs in diabetic patients was substantial especially in the lower tail of the TC distribution.  The study findings suggest that controlling of weight to reduce TC is very important in most diabetic patients, but less important in the upper tail of the TC distribution.  The quantile regression method is useful for estimating TCs at the different percentiles of the skewed TC distribution.