In order to solve the problems that traditional methods of veterinary drug named entity recognition rely on artificial design features, which is time-consuming and labor-consuming, and the amount of veterinary drug pathogenic corpus data is less in the process of building veterinary drug pathogenic knowledge graph, a method based on Att-Aux-BERT-BiLSTM-CRF of veterinary drug text named entity recognition model was proposed, which combined BERT-BiLSTM-CRF models by introducing attention mechanism and auxiliary classification layer.The text was vectorized by the BERT preprocessing model, and then connected to bi-directional long-short term memory network.The auxiliary classification mechanism was introduced, the output of the BERT layer was used as the auxiliary classification layer, and the output of the BiLSTM layer was used as the main classification layer. The attention mechanism was proposed to combine auxiliary classification layer with main classification layer to improve the overall performance.Finally, it was sent to conditional random field to construct an end-to-end deep learning model framework suitable for veterinary drug name entity recognition.In the experiment, totally 10643 sentences and 485711 characters of veterinary drug text were selected to identify four kinds of entities: drug, adverse effect, intake mode, aimal. The results showed that the model can effectively identify the entities in the veterinary drug pathogenic text, and the F1 value of recognition was 96.7%.