In order to forecast tentative equipment’s repair cost scientifically, and improve the decision-making quality of maintenance outlay, partial least squares regression (PLSR) is introduced to forecast tentative equipment’s repair cost. Aiming at the limitation of tentative equipment repair cost’s small sample, inadequate statistics, close relative eigenvector, forecasting model is constructed; based on several heavy repair data before, the heavy repair cost of special tentative equipment in use is forecasted. Meanwhile, it has the good effect to attempt optimizing the model by using variable importance in projection (VIP) in order to keep the model’s stability and improve its explaining ability and forecasting accuracy. It is proved by examples that this method can not only construct models in the case that high multi-correlation exist between variables, but also filter effectively independent variable which is of little relation to dependent variable, and simplify sample set.