In order to solve the problem of low accuracy of the demand prediction for the subsequent spare parts, a new method for the prediction of subsequent spare parts is proposed by using fuzzy soft set and Bayesian. Based on the consumption rule of the subsequent spare parts, the causal prediction model and the time series prediction model are presented, and the sum of squared residuals, the information entropy and correlation coefficient are chosen as the evaluation criterion of prediction error. The prediction effect of two single prediction methods is used as prior information. Delphi method is used to evaluate the single prediction method and construct fuzzy soft mapping. Finally, combining the prior information and the fuzzy evaluation value of the experts, the combination weight coefficient is determined by Bayesian method. An example is included to show the superiority, rationality and validity of the proposed method.