A new approach for modeling non-stationary time series was introduced in this paper.Combine the fuzzy segmentation which was proposed by Janos Abonyi with Support Vector Machines(SVMs).Firstly,a modified Support Vector Regression(SVR) was proposed;Secondly,fuzzy segment information was combined with SVR by heuristic weighting method;Thirdly,we discussed a model based on multi-SVR.Experimental results show that the method proposed in this paper has great practical values for non-stationary time series modeling.