Background: The World Health Organization (WHO) and the American Joint Cancer Committee (AJCC) modified the grading of pancreatic neuroendocrine neoplasms from a three-tier (WHO-AJCC 2010) to a four-tier system by introducing the novel category of NET G3 (WHO-AJCC 2017). Objectives: This study aims at validating the WHO-AJCC 2017 and identifying the most effective grading system. Method: A total of 2,102 patients were enrolled; entry criteria were: (i) patient underwent surgery; (ii) at least 2 years of follow-up; (iii) observation time up to 2015. Data from 34 variables were collected; grading was assessed and compared for efficacy by statistical means including Kaplan-Meier method, Cox regression analysis, Harrell’s C statistics, and Royston’s explained variation in univariable and multivariable analyses. Results: In descriptive analysis, the two grading systems demonstrated statistically significant differences for the major category sex but not for age groups. In Cox regression analysis, both grading systems showed statistically significant differences between grades for OS and EFS; however, no statistically significant difference was observed between the two G3 classes of WHO-AJCC 2017. In multivariable analysis for the two models fitted to compare efficacy, the two grading systems performed equally well with substantially similar optimal discrimination and well-explained variation for both OS and EFS. The WHO-AJCC 2017 grading system retained statistically significant difference between the two G3 classes for OS but not for EFS. Conclusions: The WHO-AJCC 2017 grading system is at least equally performing as the WHO-AJCC 2010 but allows the successful identification of the most aggressive PanNET subgroup. Grading is confirmed as probably the most powerful tool for predicting patient survival.