In this paper, a multivariate fault prognosis approach based on statistical process monitoring (SPM) methods and time series prediction for turbine machine was proposed. A principal component analysis (PCA) model using sample data under normal state was built. Firstly, fault is detected by squared prediction error (SPE) index, then predicted by AR model. With development of fault process, the SPE will produce a corresponding change and carry important fault information, so calculate statistics of SPE can be characterized and predict the trend of fault and level. A case study on the huge stack gas turbine shows the efficiency of the proposed approach.