We evaluated the clinical information value of the glycosylated hemoglobin assay by comparing it with practitioners' estimates of glucose control over the preceding 10 weeks in 216 patients with diabetes. Twenty-four per cent of the practitioners' estimates, which were based on historical and laboratory data collected during a routine office visit, differed by more than +/- 75 mg per deciliter from the actual mean blood glucose levels calculated with the glycosylated hemoglobin assay. One third of the mean blood glucose concentration fell outside the confidence intervals physicians used to bound their estimates. When examined individually or in the aggregate, historical information, such as polyuria, nocturia, or home urine testing for glucose, and laboratory information, such as fasting or random blood glucose levels, were weak predictors of the actual mean concentration of blood glucose. We conclude that the glycosylated hemoglobin assay provides information about the degree of long-term glucose control that is not otherwise obtainable in the usual clinical setting.