Abstract: According to the problem about difficult to identify the nonlinear time-varying system，a new identification method is proposed based on modified Least Squares Support Vector Machines（LSSVM）.During the algorithm’s training process，the vector base learning and automatic iterative procedures are introduced based on weighted least squares support vector machines，and a small support vector set can be obtained adaptively.Meanwhile the weights are determined by a robust method in order to reduce the effect of the outliers.Simulating for a nonlinear time-varying system，the result shows that the proposed method has a better robustness，sparseness of support vector and a real-time performance.