The invention belongs to the field of gene expression profile classification, discloses a gene expression profile distance measurement method based on deep learning, and belongs to mining and application of deep learning on biological big data. Firstly, a convolutional neural network model suitable for gene characteristic metric learning is designed to extract data characteristics, then the distance between the data is calculated by applying an improved cosine distance, and finally, the good performance of the method is measured through the classification effect of a classification algorithm.According to the method, the similarity between different gene expression profiles can be quickly and efficiently measured, and data is provided for subsequent researches such as gene classification,clustering, differential expression analysis and compound screening. Compared with a traditional gene enrichment method, the method has the advantages that the distance measurement effect between thedata is obviously improved, the manual intervention during gene expression profile analysis can be effectively reduced, the overfitting phenomenon easily generated by a conventional deep network is avoided, and the method has relatively high mobility.