In order to improve the recognition rate of vehicle in natural situations, this paper proposes a vehicle logo recognition modal based on a multi-path tree structure convolutional neural networks, which modal with different convolution kernel in the same convolutions, namely T-CNN. Firstly, different layer convolution features are obtained and are joined together as the input of the fully connected layer to get classifiers. Compared with the traditional method, the theoretical analysis and simulation results show that T-CNN can increase the recognition accuracy up to 98.43%.