The License Plate Recognition (LPR) is a major research in the fields of image processing and computer vision. The LPR is composed of License Plate Detection (LPD), License Plate Tracking (LPT) and License Plate Character Recognition (LPCR). There are databases made for the study on LPD and LPCR; however, to our best knowledge, no database is available for the study on LPT. We introduce the Multi-License Plate (MLP) database as a novel benchmark for the study on LPT. The MLP database offers 9 videos, 7592 frames, 297 s of real-world traffic scenes and 14,761 labeled bounding boxes. The variables in the MLP include lighting and shadow conditions, plate orientation, scale, resolution, partial occlusion and others. It can be one of the most challenging databases made for advancing the LPT methods. To demonstrate the use of the MLP database, we modified a few state-of-the-art approaches designed for tracking other objects and highlight the differences between LPT and other tracking tasks. The protocols and metrics to measure the LPT performance are also presented. The database is publicly available for download at https://sites.google.com/gapps.ntust.edu.tw/mlp-dataset/overview