It remains uncertain how surgeons can safely pass the learning curve of laparoscopic pancreatoduodenectomy (LPD) without potentially harming patients. We aimed to develop a difficulty scoring system (DSS) to select an appropriate patient for surgeons.A total of 773 elective pancreatoduodenectomy surgeries between July 2014 and December 2019, including 346 LPD and 427 open pancreatoduodenectomy cases, were included. A 10-level DSS for LPD was developed, and an additional 77 consecutive LPD surgeries which could provide information of the learning stage I of LPD externally validated its performance between December 2019 and December 2021.The incidences of postoperative complications (Clavien-Dindo≥III) gradually decreased from the learning curve stage I-III (20.00, 10.94, 5.79%, P =0.008, respectively). The DSS consisted of the following independent risk factors: (1) tumor location, (2) vascular resection and reconstruction, (3) learning curve stage, (4) prognostic nutritional index, (5) tumor size, and (6) benign or malignant tumor. The weighted Cohen's κ statistic of concordance between the reviewer's and calculated difficulty score index was 0.873. The C -statistics of DSS for postoperative complication (Clavien-Dindo≥III) were 0.818 in the learning curve stage I. The patients with DSS<5 had lower postoperative complications (Clavien-Dindo≥III) than those with DSS≥5 (4.35-41.18%, P =0.004) in the training cohort and had a lower postoperative pancreatic fistula (19.23-57.14%, P =0.0352), delayed gastric emptying (19.23-71.43%, P =0.001), and bile leakage rate (0.00-21.43%, P =0.0368) in validation cohort in the learning curve stage I.We developed and validated a difficulty score model for patient selection, which could facilitate the stepwise adoption of LPD for surgeons at different stages of the learning curve.Copyright © 2023 the Author(s). Published by Wolters Kluwer Health, Inc.