Purpose/Significance To improve the performance of named entity recognition (NER) model of Chinese electronic medical records (EMR) for better organization and mining of medical information. Method/Process The ERNIE-BiGRU-Attention-CRF NER model of Chinese EMR is constructed. Firstly, the ERNIE1.0 pre-training model is used to generate word vectors with semantic features, and then BiGRU is utilized to capture the global semantic features and grammatical structural features, which are fed into the Attention mechanism to further enhance the capture of the semantic features, and finally, the CRF decoding layer is connected to output the label sequences with the maximum global probability. Result/Conclusion Comparison experiments and ablation experiments are carried out on the publicly available medical text dataset CCKS2017, and examples analysis is conducted using the generated model. The model proposed in this paper achieves better recognition results.