BACKGROUND: Prior research found that in the Veterans Affairs health care system (VA), the proportion of patients adherent to oral hypoglycemic agents varies from 50% to 80% across primary care clinics. This study examined whether variation in patient and facility characteristics determined those differences. METHODS: Retrospective cohort study of 444,418 VA primary care patients with diabetes treated in 559 clinics in fiscal year (FY) 2006-2007. Patients' adherence to each oral hypoglycemic agent was computed for the first 3 months of FY2007, and averaged across agents to produce an adherence score for the patient's overall regimen. Patients with an adherence score over 0.8 were defined as adherent. Risk adjustment used hierarchical logistic regression accounting for patient factors and facility effects by clustering patients within clinics and clinics within parent VA medical centers. We then assessed the influence of risk adjustment using observed-to-expected (O/E) ratios computed for each clinic. RESULTS: The mean unadjusted proportion of adherent patients in clinics was 0.715 (interdecile range 0.559-0.826). The percent variation in patient's likelihood of being adherent explained at the patient, clinic, and parent VA medical center levels was 2.94%, 0.27%, and 0.76%, respectively. The mean clinic-level observed-to-expected ratio was 1.001 (interdecile range 0.975-1.027). CONCLUSIONS: The variation in the proportion of patients adherent across clinics remained large after risk adjustment. As patient and facility effects explained only 4% of the variance in adherence, comparing clinics based on unadjusted scores is a reasonable starting point unless more predictive patient, provider, and facility factors are identified.