Significant efforts have been made to improve physician-to-physician communication and care coordination during transition of care in order to reduce adverse events and readmissions. As electronic health records (EHRs) become widely available, many hospitals have implemented physician collaboration and hand-off tools to automatically send admission notifications, discharge summaries, and pending laboratory results to a patient's primary care physician (PCP). However, the effectiveness of such tools depends on a fundamental question that remains unstudied: who is the patient's PCP? Missing or outdated PCP information may become the bottleneck to effective patient-centered care coordination regardless of existing efforts on promoting interoperability among healthcare providers. In this paper, we characterized patient-reported PCPs and experimented with an imputation algorithm that automatically infers a patient's primary provider based on patient-provider encounter data. We compared the imputation results with patient-reported PCPs and suggested practical uses of our findings.