Objective To develop a scoring system for differential diagnosis of benign and malignant pancreatic lesions and to evaluate its diagnostic efficiency. Methods We retrospectively analyzed the medical records of patients with pancreatic lesions (with pathologically confirmed diagnosis); the patients were treated in Changhai Hospital from November 2008 to May 2013. A differential diagnosis model was created using multiple logistic regression analysis. A scoring system for differential diagnosis of benign and malignant pancreatic lesions was established based on each regression coefficient, and then was externally validated. The differential diagnosis efficiency of the scoring system was assessed by its consistency, differential ability and accuracy. Results A total of 1 000 eligible participants were included in our research. The scoring system, which was scored from 0 to 14 points, comprised 4 variables: age, anorexia, diabetes history and serum CA19-9.The system had good consistency (P=0.13), a good differential ability (area under the receiver operating characteristic [ROC] curve=0.82,95% confidence interval [CI]: 0.79-0.86,P<0.001). When score 2 was used as the predictive cut-off value, the sensitivity, specificity, accuracy, positive predictive value, negative predictive value, positive likelihood ratio and negative likelihood ratio were 81.46%, 66.88%, 77.86%, 88.26%, 54.12%, 2.46, and 0.28, respectively. The risk (88.26%) of malignant pancreatic lesions in patients with high-risk scores(>2) was significantly higher than that (45.88%) in patients with low-scores (≤2) (P<0.001). External validation results showed the scoring system had good differential ability (area under the ROC curve=0.81,95%CI: 0.76-0.86,P<0.001) and consistency (P=0.716). Conclusion The prediction model and its scoring system is of great value for risk stratification of benign and malignant pancreatic lesions, which may be serve as an initial evidence for differential diagnosis of pancreatic lesions.