Aiming at the realization of multivariate time series classification, a feature extraction method based on dimension statistics-wavelet coefficients (μσ-DWC) and a new classification model based on multi-class support vector machine (M-SVM) with tree structure were proposed. The classification algorithm was realized as follows. Firstly, map original multivariate time series to feature space by the proposed feature extraction method. The compressed representation of original time series, namely eigenvector, was obtained. Secondly, tree-structured M-SVM model was trained and constructed after getting the eigenvector representation of training set. Finally, the eigenvector of an unknown time series was extracted and put into the trained tree-structured M-SVM model. Thus, the classification process was completed and the class label of the unknown time series was obtained. The experimental results showed that the proposed algorithm provided higher classification accuracy and faster prediction speed than traditional classification methods, and gave relatively ideal training speed at the same time.