<p>For visual object tracking task, a tracked target only moves towards one of all possible directions within the search region of the current frame. However, existing one-stream transformer trackers treat all possible directions uniformly. This makes the information in the lower possible directions of the search region inevitably affect the tracking accuracy and stability. To solve this problem, we propose a direction attention based transformer tracker (DATrack), allowing the tracker to focus more on the search tokens in partial possible motion directions. Specifically, we first introduce direction tokens to provide direction information in various coarse-grained directions of the search region. After passing through the transformer backbone network, a direction token with greater correlation to the target template indicates a higher likelihood of the target moving in the corresponding direction. Then, we designed a direction attention to weight the search region tokens. In contrast to prior trackers, a search token related to the target motion direction can receive a higher attention. The search tokens weighted by direction attention are used for target localization. The experimental results show that our proposed tracker can effectively improve the robustness of tracking on multiple benchmark datasets, such as GOT-10k, LaSOT, TrackingNet, UAV123 and TNL2K.</p>