Existing similarity measure for multivariate time series can't calculate similarity accurately and rapidly. Firstly, multivariate time series are fitted with the multidimensional piecewise method. Then, average values of original points in every segment are computed as the feature pattern. Finally, inputted by feature series, dynamic time warping is used to measure the similarity of multivariate time series. The results of experiments show that the process of its parameter choice is simple, and the proposed method can guarantee the measure accuracy at relatively low computational cost.