Objective A Chinese medical named entity recognition model based on lexicon and multi-feature fusion is proposed to address the problems in existing methods including the cascading errors caused by word segmentation that affect the entity recognition effect and the issue of how to construct and fuse high-quality medical entity features. Methods Firstly the model fused the information of characters and medical vocabulary with the lexicon matching mechanism and Lattice structure. It obtained relative position embeddings by using the relative position information of words and characters and encoded Chinese character pinyin to get pinyin embeddings. Then a fusion Transformer model was proposed to mine the complementarity between different features so as to enhance the vocabulary information and promote bettter fusion of word and characyer information and pinyin information.Finally,the character representation fused with multiple festures was input into a conditional random field to obtain predicted labels.Result Experimental results on the CCKS-2019 and Resume datasets demonstrated that the proposed method achieved notable improvements across multiple evaluation metrics. Conclusion The proposed method effectively avoids the negative impact of word segmentation errors on named entity recognition. It achieves efficient fusion of various medical entity features through the fusion Transformer model. As a result it significantly enhanced the model?? s ability to recognize word boundaries thereby improving the accuracy of medical entity recognition. This provides strong support for the subsequent construction of medical knowledge graphs and the realization of intelligent medical diagnosis.