Abstract: Vulnerability classification technology is an important basis in information security vulnerability research, and is also an important means for effective management and control of vulnerability resources. In order to solve the problem of large-scale classification of vulnerabilities, an automatic vulnerability classification method was proposed based on convolutional neural network. Referring to the thought of deep learning, relevant local features of vulnerability description were acquired automatically, and the unstable problem of text training was solved through batchnorm normalized data, so as to realize the effective classification of vulnerabilities. Experiments show that compared with the traditional method, the efficiency of automatic classification of vulnerabilities can be improved to a certain degree with the proposed method.