The article proposes an intrusion detection method that bases on CNN-BiLSTM network. Firstly, the method adopts the PCA (Principal Component Analysis) method to select the features of network traffic data, which can achieve the purpose of dimensionality reduction while preserving the key features of the network traffic data. Then, CNN models and BiLSTM models were used to mine the time and spatial characteristics of network traffic data. By using the dataset to compare with other models, it is found that its accuracy, precision and recall were significantly improved, which indicating that this method has the advantages of low computational complexity，strong learning ability and high intrusion detection performance.