Network traffic classification is the basis of many network security applications and has attracted enough attention in the field of cyberspace security. Existing network traffic classification based on convolutional neural networks (CNNs) often emphasizes local patterns of traffic data while ignoring global information associations. In this paper, we propose an MLP-Mixer based multi-view multi-label neural network for network traffic classification. Compared with the existing CNN-based methods, our method adopts the MLP-Mixer structure, which is more in line with the structure of the packet than the conventional convolution operation. In our method, one packet is divided into the packet header and the packet body, together with the flow features of the packet as input from different views. We utilize a multi-label setting to learn different scenarios simultaneously to improve the classification performance by exploiting the correlations between different scenarios. Taking advantage of the above characteristics, we propose an end-to-end network traffic classification method. We conduct experiments on three public datasets, and the experimental results show that our method can achieve superior performance.