Nearly two decades ago, the National Diabetes Data Group (NDDG) and the World Health Organization (WHO) Expert Committee on Diabetes Mellitus published diagnostic criteria for diabetes. We undertook this study to compare the performance of three glycemic measures for diagnosing diabetes and to evaluate the performance of the WHO criteria. In a cross-sectional population-based sample of 1,018 Egyptians > or = 20 years of age, fasting and 2-h glucose and HbA1c levels were measured, and diabetic retinopathy was assessed by retinal photograph. Evidence for bimodal distributions was examined for each glycemic measure by fitting models for the mixture of two distributions using maximum likelihood estimates. Sensitivity and specificity for cutpoints of each glycemic measure were calculated by defining the true diabetes state (gold standard) as 1) the upper (diabetic) component of the fitted bimodal distribution for each glycemic measure, and 2) the presence of diabetic retinopathy. Receiver operating characteristic (ROC) curves were constructed to determine the performance of the glycemic measures in detecting diabetes as defined by diabetic retinopathy. In the total population, the point of intersection of the lower and upper components that minimized misclassification for the fasting and 2-h glucose and HbA1c were 7.2 mmol/l (129 mg/dl), 11.5 mmol/l (207 mg/dl), and 6.7%, respectively. When diabetic retinopathy was used to define diabetes, ROC curve analyses found that fasting and 2-h glucose values were superior to HbA1c (P < 0.01). The performance of a fasting glucose of 7.8 mmol/l (140 mg/dl) was similar to a 2-h glucose of 12.2-12.8 mmol/l (220-230 mg/dl), and the performance of a 11.1 mmol/l (200 mg/dl) 2-h glucose was similar to a fasting glucose of 6.9-7.2 mmol/l (125-130 mg/dl). Optimal cutpoints for defining diabetes differ according to how diabetes itself is defined. When diabetes is defined as the upper component of the bimodal population distribution, a fasting glucose level somewhat lower than the current WHO cutpoint and a 2-h glucose level somewhat higher than the current WHO cutpoint minimized misclassification. When diabetic retinopathy defines diabetes, we found that the current fasting diagnostic criterion favors specificity and the current 2-h criterion favors sensitivity. These results should prove valuable for defining the optimal tests and cutpoint values for diagnosing diabetes.