Yellow stem borer is one of the main pests of rice, and it can breed four complete generations in one year in Tianyang of Guangxi. By means of the insect data during the flourishing period of yellow stem borer provided by the agricultural and plant protection departments, as well as meteorological observations, adopting the stepwise regression analysis method, a meteorological early warning model of the beginning date of yellow stem borer growth peak is established. The results show that the differences between the model output and actual data for each generation are 3 days, and the historical fitting accuracy is 82.2% to 92.9%. A trial earlywarning of 2015 and 2016 was made: the difference between first generation forecast and actual situation is 3 days, and differences are 1 to 2 days for the second to fourth generations. The correction rate of historical fitting of the model and the error range of test results are satisfactory, which can be put into the application of daily service.