Recently, the correlation filter-based trackers have aroused increasing interest because of their good performance and high efficiency. A multi-scale correlation filtering tracker based on adaptive feature selection is presented. Firstly, we extract three complementary features to learn three independent filter models. By comparing the response maps, we evaluate the tracking performance of each feature, and then adaptively select the most representative feature for tracking. Secondly, to better handle occlusions and drifts, we improve the online model update strategy by setting peak response threshold as a criterion. Furthermore, we learn a separate filter model for scale estimation. The experimental results show that the proposed tracker achieves better accuracy compared with state-of-the-art correlation filter-based trackers and other popular trackers when running at 53.12 frame/s.