Abstract: To solve the problem of few training samples in modeling the path of performance degradation,the regression principle of support vector machines(SVM) based on the statistic study theory is studied.Based on the support vector machine regression(SVR) model,the methods of modeling the degradation path,lifetime prediction and reliability assessment are presented.Two kinds of performance degradation path models,single SVR model and weighted SVR model,are proposed.The example analysis indicates that the precisions of the presented models are higher than the radial basics function neural network.Specially,the weighted SVR model can be used to predict lifetime in early time,thus shortening the test time and saving outlay.