Abstract: A discriminant prediction model of time series is presented by using of multiple linear discriminant and linear autoreression models of time series. There are two selection rules of the discriminative predictor,that is (1) the stepwise induction method by schedular order; (2) the choice method by schedular non-order,and both types of selection rules may be used in this model. Thus,this discriminant prediction model may be applied to the digital records of weather time series and this model is effective on the time series under certain sutocorrelation structure.