This paper discusses tests of glycemia for the diagnosis of type 2 diabetes mellitus, with particular reference to the 1997 diagnostic criteria of the American Diabetes Association. The potential benefits of the lower diagnostic threshold for fasting plasma glucose are not well defined. However, the change in the diagnostic cut-off for diabetes mellitus affects as many as 1.9 million persons in the United States; therefore, the medical and social costs of the lower threshold may be considerable. Type 2 diabetes mellitus is defined by a threshold imposed on the continuous distribution of glycemic levels, typically with respect to risk for microvascular complications. However, the burden of type 2 diabetes relates more to macrovascular than microvascular complications. Because no clear threshold exists for macrovascular complications, a formal balancing of direct and indirect costs with both microvascular and macrovascular complications may be appropriate to establish glycemic thresholds. Because fasting plasma glucose, hemoglobin A1c, and the oral glucose tolerance test all predict diabetic complications yet test reliability is better for fasting plasma glucose and hemoglobin A1c than for the oral glucose tolerance test, we suggest an alternative diagnostic approach: If random plasma glucose is elevated (> or =11.1 mmol/L [200 mg/dL]) and the hemoglobin A1c level is more than 2 SDs above the laboratory mean, then diabetes mellitus should be diagnosed, and management should be based on the hemoglobin A1c level. If the result of only one of these tests is positive, then fasting plasma glucose should be tested to evaluate the patient for impaired fasting glucose and diabetes mellitus. The glycemic threshold for type 2 diabetes should be established by cost-effectiveness analysis. The clinical diagnosis of diabetes mellitus could be streamlined by incorporation of hemoglobin A1c into established criteria.