Purpose/Significance To construct a general medical knowledge graph covering evidence-based medical knowledge and electronic medical record (EMR) data, so as to improve the application effect of the graph. Method/Process The multi-source heterogeneous data are sorted out, the well-known knowledge graphs at home and abroad are integrated, the schema of the graphs is designed. The word embedding of RoBERTa pre-trained model is used to extract entities and relationships from medical literatures, network literatures, textbooks, medical databases and EMRs. The rule-based SWIQA framework and the manual audit strategy based on random sampling are used to evaluate the quality of the graph.Result/Conclusion A total of 128 ontologies and 1 108 relationships are identified and stored in the database in the form of triples. The semantic accuracy of the atlas is 93.8% by evaluation. The general medical knowledge graph not only covers evidence-based medical knowledge, but also includes expert experience generated in the clinical real world, which can provide support for the further development of medical AI applications.