In view of the accuracy and real-time performance improvement requirements for traffic sign recognition in natural scenes, an improved traffic sign recognition algorithm is proposed based on a multi-scale convolutional neural network (CNN). At first, the comparison experiments on image enhancement methods is carried out, contrast limited adaptive histogram equalization method is chosen as the preprocessing method to improve the image quality. Then, one kind of multi-scale CNN model is proposed to extract global and local features of the traffic sign images. Finally, the traffic signs are recognized after the combined multi-scale features are put into a fully connected SoftMax classifier. The effectiveness of the proposed algorithm is examined on the benchmark dataset---the German Traffic Sign Recognition Benchmark (GTSRB). The examination results show that the proposed algorithm can achieve 98.82% recognition accuracy and real-time processing with 0.1 ms per image on GTSRB dataset, which verified its superiority to some extent.