Multiple Object Tracking (MOT) is widely investigated in computer vision with many applications. Tracking-By-Detection (TBD) is a popular multiple-object tracking paradigm. TBD consists of the first step of object detection and the subsequent of data association, tracklet generation, and update. We propose a Similarity Learning Module (SLM) motivated from the Siamese network to extract important object appearance features and a procedure to combine object motion and appearance features effectively. This design strengthens the modeling of object motion and appearance features for data association. We design a Similarity Matching Cascade (SMC) for the data association of our SMILEtrack tracker. SMILEtrack achieves 81.06 MOTA and 80.5 IDF1 on the MOTChallenge and the MOT17 test set, respectively.