The Swedish National Patient Register (NPR) includes population-level longitudinal data, and determining the validity of NPR diagnoses is critical to undergirding the research and policy recommendations they inform. Sweden also has the integrated "Riksät" and "Stepwise" National Quality Registers (QR), with data from specialized eating disorder (ED) treatment based on structured, valid assessment methods. To validate NPR ED diagnoses, we compared ICD-10-based anorexia nervosa (AN), bulimia nervosa (BN), and unspecified ED in NPR to DSM-IV-based AN, BN, and ED not otherwise specified category (EDNOS) in QR. Patients' first diagnoses registered in QR between February 2008 and August 2013 were compared with NPR diagnoses entered within ±1 month (N = 2074). QR registration includes the semi-structured DSM-IV-based Structured ED Interview. Each ED diagnosis was analyzed separately for degree of match using several indices: overall agreement, sensitivity, positive predictive value, specificity, negative predictive value, area under the curve, and Cohen's kappa. Results showed moderate to excellent agreement depending on estimate (e.g. positive predictive values AN: 0.747; BN:.836; EDNOS: 0.761), except for a somewhat low sensitivity for BN, and EDNOS agreement was overall the lowest. Case prevalence in the NPR and QR was highly similar for AN, and within five percentage points for BN and EDNOS. Generalizability is hampered by limited age range and diagnostic resolution as well as few males. Available data precluded study of presence/absence of ED, and complementary approaches are considered for future research. We conclude that NPR ED diagnoses have acceptable validity and are appropriate for use in research.