Abstract: To improve the prediction accuracy of mildew rate on warehouse tobacco, an HP-Elman-LSSVM-based model was established. In the model, input variables are environment temperature, humidity and moisture content of tobacco, which are the major factors affecting mildew of warehouse tobacco. The training and validation samples are from the actual production data of a tobacco enterprise.The prediction model was able to efficiently predict the mildew rate. The experimental results showed that the prediction accuracy of the HP-Elman-LSSVM model is better than that of the single models. Meanwhile, by inputting different training samples, the results showed that the average relative error were between 5% and 6.5%, which can meet the requirements of engineering application.