Abstract: In order to overcome the problem of dictionary mismatch in the compressive sensing based device-free localization(DFL), a link selection learning algorithm (LSL) is proposed to enhance the DFL performance. Because the traditional shadowing-based dictionary cannot correctly describe the relationship between the received signal strength (RSS) and the target position, our algorithm first utilizes the dictionary learning (DL) technique to update the dictionary in the training phase, and uses the updated dictionary as the weight in the subsequent positioning stage. In the process of updating the dictionary, the algorithm not only reduces the dictionary dimension to speed up the dictionary learning process and improve the real-time computing speed of the algorithm, but also filters out the outlier links by selecting those links through the confidence region. The indoor and outdoor experimental results show that the proposed method can effectively mitigate the model error and improve the positioning accuracy with the low computational complexity.