A bstract: Objective To establish a risk model for predicting the common pathological classification of lung cancer by the combination of six serum tumor markers. Methods The clinical data of 342 patients with lung cancer and 91 patients with suspicious lung imaging and diagnosed as non-lung cancer, who were treated in the Affiliated Cancer Hospital of X injiang Medical University during May 2012 and May 2013, were analyzed retrospectively. The relationship between tumor markers and pathological subtypes of lung cancer was obtained by the heat map analysis of 6 tumor markers. The independent predictive indicators of each pathological subtype were screened by the Logistic regression analysis, and a prediction model was constructed. The predictive ability of each model was evaluated by the area under the ROC curve (AUCROC ). The Bootstrap resampling method was used to validate the predictive ability of each model. Results The heat map analysis showed that the combination of pro-GRP and NSE was better in predicting small cell lung cancer, SCC and CY-FRA21-1 in predicting lung squamous cell carcinoma, and CA125 and CEA in predicting lung adenocarcinoma. The areas under the ROC curves of prediction models for small-cell lung cancer, lung squamous carcinoma and lung adenocarcinoma were 0.938, 0.965 and .0.965, respectively. Internal validation of the predictive ability of the three models showed that their predictive effects were ideal. Conclusion The risk prediction models for pathological lassifcation of lung cancer patients constucted by serum tumor markers combined with general data of patients have good predictive ffects. After further venification in the future, they could be displayed on the test report to provide help for clinicians in further diagnosis and treatment.