It is the scientific basis to slow down the adverse effect of flood,drought and so on to be able to obtain accurate prediction information. Under the background of global warming, many kinds of meteorological disasters caused by the regional extreme weather events occur frequently,and bring many new challenges on the economic and social development. The climatic prediction is a hard technical problem all over the world. When we build the season prediction models,if considering the influence of the decadal trend background,can we improve the prediction level? This problem is worth further researching. Using precipitation data in winter at 33 meteorological stations in Xinjiang and 108 previous circulation characteristic indices, considering their linear trend, two selecting decadal background trend schemes of Original sequence" and Filter linear trend" are designed. Using "running correlation ,stepwise regression and ensemble analysis" method, with the precipitation as prediction object and the circulation characteristic indices as predictors, the multiple regression prediction models with two different selecting decadal background trend schemes at 33 stations are founded in respectively. The results show that the ACC of the two decadal background trend schemes are both above 0.238, and hold a certain predictive ability. In contrast, the prediction of Filter linear trend" is better than that of "Original sequence". With correlation coefficient independent test as indicators, the "Ensemble analysis" prediction is obtained. The prediction is superior to the scheme of "Original sequence" , compared with the scheme of "Filter linear trend" to improve slightly. Considering the decadal background trend of winter precipitation, we can improve the prediction in Xinjiang.