Purpose: Annual foot examinations (FE) constitute a critical component of care for patients with diabetes. Documented evidence of FE is central to quality of care reporting; however, manual abstraction of electronic medical records (EMR) is slow, expensive, and subject to error. Our objective is to test the hypothesis that automatic processing of the language in the EMR results in ascertaining FE evidence with accuracy comparable to manual abstraction. Mehtods: The text of in- and out-patient clinical reports was searched with three natural language (NL) queries for evidence of the neurological, vascular and structural components of FE. The components were defined as a set of NL terms derived partly from the clinical practice guidelines for FE and partly from the text of clinical reports with known FE status. The queries found patients that had evidence of: a) at least one of the three components; b) any two of three components; c) all three components. Validation was performed by manual abstraction of medical records. A reference standard consisted of three independent sets: a development (n=200) and validation set (n=118) for the NL queries, and a reliability set (n=80) used to determine the reliability of the manual abstraction (manually re-audited and the percent agreement between the re-audit and the initial audit used as a measure of reliability). Results: The reliability of manual abstraction for one of three components was 91% (95%CI: 85-97). The accuracy of the natural language query requiring one of three components was 89% (95%CI: 83-95) with 93% (95%CI: 88-98) sensitivity and 71% (95%CI: 52-90) specificity. The accuracy of the query requiring any two of three components was 88% (95%CI: 82-94) with 91% (95%CI: 85-96) sensitivity and 76% (95%CI: 58-94) specificity. The accuracy of the query requiring all three components was 75% (95%CI: 68-83) with 75% (95%CI: 67-84) sensitivity and 76% (95%CI: 58-94) specificity. Conclusion: This study is the first to show that querying the NL of the EMR offers accuracy of FE ascertainment comparable to manual abstraction and therefore may be used proactively for monitoring compliance with quality of care guidelines. Our low cost and robust methodology is scalable to monitoring large numbers of patients to improve quality of care and, consequently, outcomes.