Problems in using medical records to assess outcomes of diabetes care have not been well defined. We reviewed the medical records of 288 patients with diabetes receiving ambulatory care over a 2-year period. We determined the availability of different tests of glycemic control and described site performance as the percentage of patients with a blood glucose exceeding either 180 or 240 mg/dl. Glycosylated hemoglobin determinations were performed in only 26.7% of patients. A blood glucose was available in 208 patients (72.2%) during a 6-month outcome period. For almost 50% of the sample, the glucose was greater than 180 mg/dl, whereas in 20% it exceeded 240 mg/dl. Judgments of whether sites differed in performance depended on how control was defined. Using a single glucose determination and a threshold of 180 mg/dl, similar fractions of patients were poorly controlled at each site (51.2 versus 45.0 versus 47.0%) (P = 0.75). At 240 mg/dl, although, one site performed much worse than the other two (14.6 versus 16.7 versus 31.8%) (P = 0.02). These results highlight difficulties in defining the outcome measure when using medical records to evaluate quality of care.