In order to improve the prediction accuracy of the occurrence severity degree of the Dendrolimus punctatus larvae and find simple and accurate forecasting method. The method of stationary time series, regression?forecast, Markov chains, BP neural network and contingency table?analysis were applied to establish the prediction model of the occurrence severity degree of overwintering generation, the first generation and the second generation Dendrolimus punctatus larvae from 1983 to 2014 in Qianshan county of Anhui Province and it was used to study the historical coincidence rate, and then the predicted result was proved with the actual happening situaion in 2015 and 2016. Result shows that: The predicted results which used the more simple calculation methods of stationary time series and contingency table were accurrte; BP neural network and Markov chain method to predict the result were very accurate. The predicted result of single regression model that the contemporary amount of eggs from the egg stage predicted the occurence severity degree of the contemporary larvae was very accurate in the regression model and other predicted result with the single regression mode was a bit poor, so multiple regression model and the stepwise regression model were better than the single regression model. BP neural network model is a kind of ideal prediction model.