Class imbalance problem is a key factor that restricts the application of machine learning technology in the field of intrusion detection. The classification result will be significantly affected if the training data is imbalanced due to the classifier’s tendency towards the majority class. A data augmentation method for intrusion detection combining generative adversarial networks and deep neural network is proposed in this paper in order to solve the problem of sample imbalance and lack of sample because of the confidentiality of the data. Finally, the model is evaluated on the NSL-KDD dataset and the result shows that our method not only has higher accuracy rate, but also has higher detection rate for unknown attacks and attack types with only a few samples compared with the traditional algorithms.