Based on the terahertz timedomain spectroscopy system, 4 kinds of rubber samples were detected. Comparing with PCA and CCA, Kernel principal component analysis (KPCA) and kernel canonical correlation analysis (KCCA) were carried out on the feature extraction of rubber terahertz spectrum. The classification model was established by support vector machine (SVM) to classify the rubber samples. Finally, the recognition results of partial least squares (PLSDA) are used as the reference. The experimental results show that SVM can be used to classify the spectrum of rubber combined with the feature extraction methods. The classification effect of KPCASVM on the absorption spectrum is the best, and PLSDA is better than SVM on refraction spectrum classification. Meanwhile, KPCA is better than the standard KCCA method for the feature extraction of the spectrum. The experiment provides a new method for the identification and analysis of rubber.