This study aims to use natural language processing (NLP) of electronic health information to verify the accuracy of diagnoses of anorexia nervosa (AN) and bulimia nervosa (BN), a task which is typically done via manual chart review, which can be time-consuming and vulnerable to human error. Electronic health records of 400 individuals attending hospitals in the Region of Southern Denmark with ICD diagnoses of AN (F50.0), atypical AN (F50.1), BN (F50.2), or atypical BN (F50.3) were selected. For each individual the chart was reviewed manually and by a NLP model to verify whether an individual met criteria for the condition they were diagnosed with. The NLP model was used to identify certain keywords in medical charts that are related to specific eating disorder (ED) diagnoses. Positive predictive value (PPV), which calculates the likelihood that someone who tests positive actually has a condition, was calculated for each diagnostic code. Overall, 93% of diagnosis codes were assigned correctly using manual chart review assisted by an NLP model. PPV was highest for BN and atypical AN at 96% and lowest for AN and atypical BN at 90%. In cases where an incorrect diagnosis was given, the most common reason was that a patient more closely fit criteria for a different ED. Overall, this study finds that the diagnosis codes of F50.0, F50.1, F50.2, and F50.3 have high validity in the Danish hospital system. The accuracy of ED diagnoses is somewhat higher than for other mental health conditions. Using the assistance of an NLP model, the authors were able to reduce the risk of overlooking important clinical information.