Gross agricultural product is an important indication to measure the agricultural development level of a region. It would be affected by many factors, having the characteristics of non-linearity. For this reason, LM-BP neural network was put forward as the model and method for predicting gross agricultural product. Taking the indications of the sown area of crop, the output of grain, sugarcane, cassava, tea, meat, aquatic products, turpentine and oil-tea camellia seed, etc. as inputs, during 2000 to 2012 in Guangxi, the gross agricultural product data from the analysis of simulation experiment show that the prediction of LM-BP neural network fits well with actual results.