In order to improve the recognition performance of intrusion detection, a graph convolutional neural network for multiple classification intrusion detection is proposed, named as GCNID. Firstly, the detection data are preprocessed by numerical and normalized, so that the disadvantage effect of numerical differences among various features can be reduced. Then, the adjacency matrix for intrusion data in GCIND is constructed by the k-nearest neighbor method, which utilizes the Euclide distance between different features as evaluation criterion. Besides, the adjacency matrix and the modified detection data are set as the input of GCNID framework. Furthermore, the neighbors information of each detection data are added into itself by the graph convolutional layer, such that more effective intrusion features can be extracted by GCNID. What s more, the parameters in the proposed method are optimized by back propagation until the network convergence. Finally, the Softmax logistic regression model is used to classify the types of intrusion. Simulation results show that the proposed method can not only improve the detection accuracy, but also classify the unknown attack types. The performance shows that the GCNID has a good ability of accuracy and generalization.