BACKGROUND It is unclear why the prevalence of diabetes and prediabetes, especially prediabetes, between diagnosed by oral glucose tolerance test (OGTT) and hemoglobin A1c (HbA1c) criteria, is substantially discordant. OBJECTIVE We aimed to evaluate the effects of obesity on the agreement between HbA1c and OGTT for diagnosing diabetes and prediabetes and identify the optimal HbA1c cutoff values in different body mass index (BMI) classifications. METHODS In a population-based, cross-sectional study in Harbin, China, 4325 individuals aged 20-74 years without a prior diagnosed diabetes were involved in this study. RESULTS measure The performance and optimal cutoff points of HbA1c were assessed by receiver-operating characteristic curve. The contribution of BMI to HbA1c was analyzed by structural equational model. RESULTS The agreement between HbA1c criteria and OGTT decreased with BMI gain (κ = 0.359, 0.312, and 0.275 in a normal weight, overweight, and obese population, respectively). The structural equational model results showed that BMI was significantly associated with HbA1c in normal glucose tolerance and prediabetes subjects but not in diabetes subjects. At a specificity of 80% for prediabetes and 97.5% for diabetes, the optimal HbA1c cutoff points for prediabetes and diabetes were 5.6% and 6.4% in normal-weight, 5.7% and 6.5% in overweight, and 6.0% and 6.5% in an obese population. When the new HbA1c cutoff values were used, the agreement in obese subjects increased almost to the level in normal-weight subjects. CONCLUSIONS The poor agreement between HbA1c and OGTT criteria in an obese population can be significantly improved through increasing the HbA1c threshold for prediabetes.