Objective To investigate the potential feasibility of lung cancer diagnosis with saliva surface-enhanced Raman spectroscopy, and to obtain the relatively optimal diagnosis model of lung cancer by data mining.Methods In this paper, saliva samples of 18 healthy individuals and 59 lung cancer patients were measured and analyzed the spectra by portable SERS detection system.We established the support vector machine (SVM) and random forests by data mining technology, compared with traditional Fisher discriminant model, and then discussed the auxiliary diagnosis efficiency for lung cancer with the models.Results The diagnosis indexes of the SVM and random forest were higher than Fisher discriminant analysis.We considered SVM and random forest were the optimal classification models for the diagnosis of lung cancer.Conclusions The results showed that the study of surface enhanced Raman spectroscopy based on data mining might be a new type tool for the diagnosis of lung cancer.