Most existing convolutional neural network-based text classification methods have problems such as inadequate feature extraction, long-distance dependency. This study presents a novel deep multi-feature fusion neural network for text classification. The convolutional neural network is employed to extract the local features of the text. An improved encoder with self-attention method is proposed to build the global relationships. Furthermore, a multiple feature fusion strategy is provided to integrate the features learned from the CNN and the encoder, which is beneficial to excavating the potential semantic information in a global perspective. The experimental results demonstrate that our proposed method has better classification performance than the alternative approaches.