BACKGROUND:  Use of electronic healthcare databases to examine pharmaceutical use and effects faces limitations inherent in the source of data.  Medical records document prescribing but not consumption of medication; administrative databases document dispensing but not the original intent of the prescriber. OBJECTIVES: To evaluate a patient-level linked database of electronic medical records (EMR) and administrative claim for use in measuring both prescribing and dispensing of pharmaceuticals. METHODS: Claims data from Thomson Reuters MarketScan database were linked to the GE Centricity EMR database using probabilistic methods to overcome the de-identification required of both databases under US privacy laws.  Patient-level records were matched based on demographic characteristics and calendar dates of physician visits.  Multiple visits were required to reduce the likelihood of mismatches.  The agreement between prescribing records in the EMR and claims for dispensing of medication was explored in the context of self-administered medication for osteoporosis, including bisphosphonates and raloxifene. RESULTS: Using data from 2004-2009, 219,529 patients were matched between the two data sources.  Mean age was 43 years and 57% were female.  We identified 2,331 patients whose medical record showed new prescriptions for bisphosphonates or raloxifene (no evidence of use in the prior 180 days).  Pharmacy claims indicated these prescriptions were filled by 56% of patients within 7 days, 75% within 30 days, and 86% within 90 days. CONCLUSIONS: It is to be expected that there will be some degree of noncompliance, hence incomplete filling of prescriptions, as observed.  The lag from dispensing to prescribing among many patients is also consistent with use of samples for initial prescriptions.  Overall, the example of osteoporosis therapy shows consistency between prescribing in EMR data and dispensing in claims data among a sample of probabilistically linked patient records.