We estimate doctor value-added (VA) combining population-wide patient-doctor register data with exogenous variation in the assignment of patients to GPs. We find substantial variation in the quality of GPs as measured by patients' post-assignment mortality. Certain doctor characteristics and practice styles predict VA, but most of the variation in GP quality is driven by differences in GPs' unobserved ability. VA variation across GPs primarily reflects differences in GPs' ability to engage in prevention and assign the right procedure to the patient. Finally, patients are unable to identify who the high-quality doctors are, and patient-generated ratings are uncorrelated with GP VA.