Background/Aims: Most HMORN members offer both insurance coverage and health care. Health plan enrollment provides a well-defined population denominator for HMORN-based research. The HMORN common data model combines electronic data, routinely collected in health care delivery or claims processing, into the Virtual Data Warehouse (VDW). Sites with health plans capture insurance enrollment in the VDW enrollment file. Recently HMORN included sites that deliver health care without offering insurance. Since health plan membership is unknown, population denominators must be determined using alternate methods. This study validated a utilization-based proxy enrollment (PE) using standard VDW enrollment (SE) in 5 HMORN sites with Epic EHR and SE files. Methods: The utilization-based algorithm defines PE start at first of two non-ancillary health care visits (at least one being a primary care visit) separated by at least 90 days. PE ends at death or at the last qualifying visit if there is no utilization in the following 3-year period. PE files were built at each site by applying a utilization-based algorithm to base tables created from Clarity. The PE and SE extracts included study ID, age, gender and the start/end of enrollment periods between 2000 and 2012, based on availability of Clarity data at the sites. The agreement between PE and SE was evaluated using differences in start/end of the first enrollment periods (days). Results: The differences between PE and SE starts varied by site with greater variation in children (<20 years) and older adults (64+ years). The differences varied by gender, but the differences between the genders were smaller in the young and the old. The differences were larger for males between ages of ~16 and 64, indicating less utilization by males than by females. The differences between PE and SE ends were generally negative across ages, indicating that PE extended beyond SE. Gender differences between PE and SE starts were similar to differences between PE and SE ends. These empirical results were confirmed by multivariate regression modeling. Conclusions: Agreement between PE and SE could be improved using additional parameters, as well as possible adjustment of the time lag for PE end in the utilization-based algorithm.