Abstract: The level of operation of agricultural mechanization is a complicated non-linear system, whose developmental changes have dual trends of increase and fluctuation in Heilongjiang Province, and has a high request to the fitting method. Through conducting the research on prediction method for operation level of agricultural mechanization, this study established basic prediction models based on tradition prediction models such as GM(1,1), smoothness and regression. The authors also combined the merits of tradition prediction models and BP neural network to establish comprehensive models, such as serial grey neural network, serial smoothness neural network, etc. The operation level of agricultural mechanization such as ploughing, sowing, harvesting, plant protection and irrigation in Heilongjiang Province from 2008 to 2015 were predicted. The results validate the effectiveness of the proposed models, and provide a new method for predicting the operation level of agricultural mechanization; mechanization level of ploughing, sowing, harvesting are higher than the other production links, irrigation is the main bottleneck, and needs to be further enhanced