OBJECTIVES: To assess the prevalence of potentially undiagnosed diabetes mellitus (UDM) in a nationally representative patient sample. METHODS: The data source was GE’s Medical Quality Improvement Consortium (MQIC) database (February 2009) containing electronic medical record (EMR) data on >11 million patients in the U.S. Two previously published (Holt, et al, 2008) search strategies were applied to identify patients without diagnosis or medication evidence of DM, but with a fasting (FBG) or random blood glucose (RBG) result available (denominator for prevalence estimates). Strategy A patients were non-DM patients whose last glucose on record (LGOR) = RBG ≥11.1 or FBG ≥7.0 mmol/l. Strategy B patients were non-DM patients with LGOR (FBG or RBG) ≥7.0 mmol/l.  Strategy A and B patients were each grouped by age/sex categories, and prevalence of UDM calculated. The time since LGOR to datacut date was also assessed. RESULTS: From 11,196,881 total patients, 923,007 had diagnosed DM on record (n= 570,723) or were presumed to have DM on the basis of prescribed oral hypoglycemics or insulin (n=352,284). After excluding additional patients with gestational diabetes and impaired glucose tolerance/prediabetes,10,147,355 remained.  Of these, 3,799,599 had a glucose result available, with 38,068 identified as possible UDM using Strategy A (prevalence 0.38%), and 221,624 using Strategy B (prevalence 2.18%).  In both instances, UDM prevalence increased with increasing age, in both sexes. Over 2 years had elapsed since the LGOR for over 50% of Strategy A patients, and 40% of Strategy B patients.  CONCLUSIONS: The application of simple search algorithms to a large EMR database suggests there may be substantial underdiagnosis of DM in the US general population.