Objective To investigate the minimal and essential practice for learning breast imaging reporting and data system (BI-RADS ) for ultrasound, investigate the regular pattern of its learning, so as to provide reference for clinical training and evaluation. Methods Eight first-year residents, ten second-year residents and ten refresher doctors were trained for BI-RADS for ultrasound using the non-randomized experimental studies from October 2017 to October 2018. One hundred breast tumors were randomly divided into 5 groups with 20 cases in each group. Scores were compared among the 5 groups according to the results of chief physicians and the learning curve of was analyzed. Results The learning curve for the three categories of physicians increased significantly from the second time, while the trend of the second time to the fifth time was relatively stable. There were statistically significant differences in the diagnostic scores between the first time and second time for first-year residents, second-year residents and refresher doctors [(28.8±16.9) vs (50.0±9.3), (31.5±7.5) vs (53.5±6.7), (50.5±8.0) vs (62.0±13.4); all P＜0.05]. Conclusions Residents and refresher physicians may master the BI-RADS dictionary after the theoretical training of BI-RADS dictionary and the reading training of about 40 cases.