INTRODUCTION: We studied variance in glycated hemoglobin (HbA1c) values among adults with diabetes to identify variation in quality of diabetes care at the levels of patient, physician, and clinic, and to identify which levels contribute the most to variation and which variables at each level are related to quality of diabetes care. METHODS: Study subjects were 120 primary care physicians and their 2589 eligible adult patients with diabetes seen at 18 clinics. The dependent variable was HbA1c values recorded in clinical databases. Multivariate hierarchical models were used to partition variation in HbA1c values across the levels of patient, physician, or clinic and to identify significant predictors of HbA1c at each level. RESULTS: More than 95% of variance in HbA1c values was attributable to the patient level. Much less variance was seen at the physician and clinic level. Inclusion of patient and physician covariates did not substantially change this pattern of results. Intensification of pharmacotherapy (t = -7.40, P < .01) and patient age (t = 2.10, P < .05) were related to favorable change in HbA1c. Physician age, physician specialty, number of diabetes patients per physician, patient comorbidity, and clinic assignment did not predict change in HbA1c value. The overall model with covariates explained 11.8% of change in HbA1c value over time. CONCLUSION: These data suggest that most variance in HbA1c values is attributable to patient factors, although physicians play a major role in some patient factors (e.g., intensification of medication). These findings may lead to more effective care-improvement strategies and accountability measures.