Endoscopic retrograde cholangiopancreatography (ERCP) is a common diagnostic method and therapeutic technique for biliary and pancreatic diseases in clinical practice. In order to shorten the training period and reduce the occurrence of surgical complications due to operational issues, operators need extensive training through training models to reach the level of independently performing a procedure. Here, the authors summarize and analyze the current status and existing problems of different models used at home and abroad: mechanical simulation operation models and computer simulation models have characteristics such as flexibility and visibility, suitable for beginners to understand the procedural flow, but lack realism compared to other models; live animal models have the advantage of complete operational training and management of complications, but they are relatively expensive and do not fully match the human anatomy for training purposes; ex vivo animal models have the advantages of low cost and better realism, suitable for advanced learning and initial training in managing complications. Finally, based on the characteristics of different models, the authors innovatively propose a phased joint ERCP training scheme—the five-stage method, and also suggest new directions for the iterative development of ERCP training models.