This paper introduces new performance measures and a large-scale, fully-annotated, calibrated multi-camera tracking dataset to advance the field of multi-target, multi-camera (MTMC) tracking. The proposed measures, called ID precision and ID recall, focus on correctly identifying identities rather than penalizing specific types of errors. The dataset, DukeMTMC, consists of over 2 million frames from 8 cameras, covering more than 2,700 identities over 85 minutes, and is the largest and most realistic dataset for MTMC tracking to date. The paper also presents a reference system that extends a single-camera tracking system to the multi-camera setting, demonstrating comparable performance to state-of-the-art methods. The authors show that their measures accurately reflect bottom-line identity match performance in multi-camera settings, and that the dataset poses realistic challenges for current trackers.This paper introduces new performance measures and a large-scale, fully-annotated, calibrated multi-camera tracking dataset to advance the field of multi-target, multi-camera (MTMC) tracking. The proposed measures, called ID precision and ID recall, focus on correctly identifying identities rather than penalizing specific types of errors. The dataset, DukeMTMC, consists of over 2 million frames from 8 cameras, covering more than 2,700 identities over 85 minutes, and is the largest and most realistic dataset for MTMC tracking to date. The paper also presents a reference system that extends a single-camera tracking system to the multi-camera setting, demonstrating comparable performance to state-of-the-art methods. The authors show that their measures accurately reflect bottom-line identity match performance in multi-camera settings, and that the dataset poses realistic challenges for current trackers.