OBJECTIVES: To examine the concordance in diabetes diagnosis between electronic medical records (EMR) and claims data and to explore an appropriate definition for diabetes in EMR data. METHODS: Retrospective study using the Linked Quintile EMR and MarketScan Commercial and Medicare Databases (MSN).  Patients with at least one EMR every year and continuous enrollment in MSN between January 1, 2009 through December 31, 2011 were included in the study. Patients with at least one inpatient or two outpatient claims for diabetes (ICD-9-CM 250.xx) in MSN were identified as diabetic patients. Diabetes was defined in the EMR data using various criteria, including HbA1C≥6.5%, diabetes in the problem list, any insulin prescription, ≥2 oral anti-diabetic agents, and ≥2 abnormal fasting plasma glucose tests.  An extended definition of diabetes based on the presence of any of the criteria mentioned above was also evaluated. RESULTS: Among the 257,899 patients meeting the study inclusion criteria, 34,383 were identified as diabetic patients using MSN claims.  When the extended definition was used, 27,219 patients were defined as diabetic in EMR, an 11.8% increase compared with those identified using the HbA1C≥6.5% criterion, and 2.35 times of those with a diabetes diagnosis in the problem list. Assuming claims to be the gold standard, the extended definition also had a higher sensitivity at 79.2% (versus 70.8% when HbA1c≥6.5% criterion was used alone and 33.6% when the criterion of diabetes on problem list was used alone). The specificity of the extended definition was 97.9%. CONCLUSIONS: This study explored the feasibility of identifying the diabetic patients in the Quintile EMR data and confirmed good concordance of the diabetes definition between EMRs and claims. When linked EMR and claims data are used, the diagnosis of diabetes is strengthened when patients have evidence of diabetes in both EMR and claims.