A new data mining method of Support Vector Machines (SVM) is applied on the classification of rock mass in tunnels. SVM is a novel powerful leaning method that based on Statistical Learning Theory. SVM can solve small-sample learning problems better than neural network. Data samples from Niba Mountains tunnel are used to train the SVM with different kernels. The mapping relationship between judge factors and rock mass classes is used. The SVM can discriminate and provide class-unknown data samples of rock mass. The result of the classification shows that SVM with polynomial kernel has a high accuracy. So this is an intelligent classification of rock mass method that can be applied to classify rock mass in tunnels.