Abstract: Objective：To distinguish lung cancer based on 6 tumor markers of lung cancer and 4 epigenetic tumor markers combined with classifying models of decision tree C5.0(CART C5.0)， and develop the best model for the diagnosis of lung cancer. Methods：The levels of serum CEA， gastrin， NSE， sialic acid (SA)， Ca， Cu/Zn， DNMT1， DNMT3A， DNMT3B， HDAC1 in 180 patients with lung cancer and 243 patients with lung benign disease were detected by means of radioimmunology， double antibody sandwich method， modified resorcinol chromogenic method， fully automated analyzer， atomic absorption spectrophotometry， enzyme linked immunosorbent assay (ELISA)， respectively， and then developed CART C5.0 models. Results： The sensitivities of 6 tumor markers， 4 epigenetic tumor markers and 10 tumor markers in the CART C5.0 model were 0.383， 0.378， 0.944， respectively.The specificities were 0.905， 0.864， 0.934， respectively.The accuracies were 0.683， 0.657， 0.939.The areas under receiver operating curve(AUC)were 0.644， 0.621， 0.939， respectively. The sensitivity， specificity， accuracy and AUC of 10 tumor markers were over 90%， and better than the sensitivities of 6 tumor markers， 4 epigenetic tumor markers models， and there were statistical difference (P<0.05). Conclusion： The CART C5.0 model based on 10 tumor markers is obviously superior to that of the 6 biomarkers and 4 epigenetic markers， which improved the diagnosis rate of lung cancer.