Objective To investigate the learning curve for da Vinci robot-assisted mediastinal tumor resection (DRMTR). Methods A total of 50 consecutive patients received DRMTR between March 2011 and September 2012 in our hospital. Clinical data of the 50 patients were collected and analyzed. There were 23 males, 27 females aged 46.9(17–80) years. The learning curve was evaluated by using the cumulative sum (CUSUM) analysis. Results The mean operation time was 124.6 min. The CUSUM learning curve was best modeled as a third-order polynomial curve with the equation: CUSUM=0.046×case-number3–4.681×case-number2+127.508×case-number–237.940, which had a highR2 value of 0.868. The fitting curve reached the top after the 19th case, which suggested that the surgeons master the technique after they finished 19 cases. As a cut-off point, the 19th case divided the learning curve into two phases, in which there was statistical diffference in operation time (P<0.01), intraoperative blood loss (P<0.01), the postoperative duration of chest tube drainage (P<0.01 ) and the rate of postoperative complications (P<0.05 ). Conclusion The DRMTR identified by CUSUM analysis represents two characteristic stages of DRMTR: the learning stage and the mastery stage. It is suggested from our data that the surgeons need finish about 19 cases to master DRMTR.