TLD is a real-time long-term tracking system that decomposes the tasks into three components: tracking, learning and detection. The learning estimates detector's errors and updates it to avoid these errors in the future. However, Current implementation of TLD trains only the detector and the tracker stay fixed. As a result, the tracker makes always the same errors. In our paper, we develop a novel training method which combines naive Bayes classifier with the optical flow based on the TLD algorithm to train the tracker. The proposed algorithm mainly consists of two stages: one stage for training and a second stage for tracking. For the training stage, we sample some positive samples near the current target location and negative samples far away from the object center to update the classifier from the current frame. For the tracking stage, Optical flow tracker estimates the object's motion between consecutive frames under the assumption that the frame-to-frame motion is limited and the object is visible. And then we sample a set of image patches, using the classifier to each patch. We determine the target with the maximal classification score. With these definitions, we conduct extensive experiments and comparisons for the proposed method. The comparisons and experiments well demonstrate the effectiveness of our work.