Based on the named entity recognition in electronic medical records is of great significance to medical treatment AI and the construction of medical knowledge graph，a proposal has been made of a named entity recognition method based on medical categories. First, the electronic medical record is to be divided into 4 categories according to the entity characteristics of the corpus of electronic medical records. Then, the feature sets are to be constructed respectively for the medical categories, followed by an identification of the named entities of such five named entities as body parts, symptoms and signs, inspection and test, disease and diagnosis, and treatment by using the conditional random field model. Finally, a comparison has been made between the recognition results based on medical class feature sets and the general feature sets. The results show that the effect of named entity recognition based on medical categories has been significantly improved, enabling it to meet the application requirement effectively.