In order to improve the prediction accuracy of pest’s occurrence quantity, this paper proposed a nonlinear prediction model for pest’s occurrence quantity based on chaotic theory. Time series for pest’s occurrence quantity were reconstructed by phase space reconstruction, and then input into the least squares support vector machine to learn and establish prediction model for the pest’s occurrence quantity, the test experiment is carried out using Dendrolimus punctatus occurrence area data in Simao, Puer and Xianju county, Zhejiang. The results show that the prediction values of Dendrolimus punctatus occurrence area were very close to the actual production, the mean absolute percent error of predicted results for Dendrolimus punctatus occurrence area in in Simao, Puer and Xianju county, Zhejiang were 0.90% and 2.44% respectively, the prediction results were better than that of BP neural network and linear prediction model.