BACKGROUND: Studies of influenza vaccination using electronic medical records rely on accurate classification of vaccination status. Vaccinations not entered into electronic records would be unavailable for study. PURPOSE: This study evaluated the sensitivity and negative predictive value (NPV) of electronic records for influenza vaccination and factors associated with failure to capture vaccinations. METHODS: In four diverse medical care organizations in the Vaccine Safety Datalink, those aged 50-79 years with no influenza vaccination record during the 2007-2008 season were surveyed by telephone, and electronic records were analyzed in 2008. The sensitivity and NPV of electronic records were estimated, using survey responses as the gold standard. Logistic regression models determined associations between 1-NPV and demographic factors, risk of influenza complications, and healthcare utilization levels. RESULTS: Data were obtained for 933 survey participants and 1,085,916 medical care organization members. Sites varied significantly in the sensitivity (51%, 68%, 79%, 89%) and NPV (46%, 62%, 66%, 87%) of electronic records. In multivariate analysis, the rate of failure to capture vaccinations was significantly higher for those aged 65-79 years than for those aged 50-64 years at three sites. Of vaccinations not captured by electronic records, 58% were reportedly administered in nontraditional settings, usually workplaces; the rest were given within the sites. CONCLUSIONS: Influenza vaccination studies relying on electronic records may misclassify substantial proportions of vaccinated individuals as unvaccinated, producing biased estimates of vaccine effectiveness. Sites with limited sensitivity to capture vaccinations administered within their organization should seek possible remedies. More complete capture of vaccinations administered to older patients and in nontraditional settings would further reduce misclassification.