Traffic Sign Recognition (TSR) is very important for driverless systems and driver assistance systems. Due to the small size of traffic signs in the wild, the traffic sign becomes very challenging. In this paper, an inception convolutional neural network is designed to solve the traffic sign classification problem. A large receptive field is generated by multiple small filters instead of a single large filter. Moreover, Inspired by Inception V3, inception block is used, which makes the combination of multiple convolution output be optimized. Thus, the coarse cue in the shallow layer and the fine cue in the deeper layer are fused to improve the visual expression capability of the model. The proposed method is evaluated on three famous traffic sign datasets: the German Traffic Sign Recognition Benchmark (GTSRB), the Swedish Traffic Signs Dataset (STSD), and the 2015 Traffic Sign Recognition Competition Dataset. The experimental results demonstrate the effectiveness and robustness of our methods.