There are many factors that affect the mean time between failures of CNC machine tool holder system, it is difficult to extract the hidden information from the known data by using a single model. A combined model based on STL time series decomposition forecasting algorithm was applied was decompose the original fault data into trend term, seasonal term and random term.Exponential smoothing method and support vector machine regression were applied to forecast the first two data respectively. The predictions were combined according to the additive model of temporal decomposition.So combined model prediction results were obtained, and the combination model was compared with single forecast model. The example shows that the combined model is superior to the single model. This method is applied to the prediction of MTBF, which is helpful for the staff to take measures in advance according to the time point of failure, and provides a new research method for the reliability assessment of CNC machine tools.