Abstract: The quantitative prediction of agricultural equipment level can provide the basis for making plan of agricultural mechanization development. ARIMA time series and BP neural network model were chosen to construct a new combination prediction model based on Shapely value method to determine each prediction model weight. By the model, the agricultural equipment level in Shanxi province were predicted according to the total power of agricultural machinery, the large medium tractors, the small tractors, the matching implements of large medium tractors, and the matching implements of small tractors from 1979 to 2005. The results show that the prediction precision of combination prediction model is higher than any selected prediction model, and is feasible and effective for the prediction of agricultural equipment level. The data of the total power of agricultural machinery, the large medium tractors, the small tractors, the matching implements of large medium tractors, and the matching implements of small tractors , the rate of implements of large medium tractors , the rate of small tractors in Shanxi province were predicted ,and will attain to 2619 million kW, 43497 sets, 297546 sets, 84683 sets, 327743 sets, 1.96, 1.15 in 2010 by combination prediction model.