Abstract: As food security is the most important thing in governing the country, it is of great significance to grasp the change of grain moisture content in real time for food security storage. This article choose late indica rice “Zhongzhe You No. 8” as the research object, under the condition of different temperature and different humidity, get grain moisture content measurement by NIR experiments, and the BP neural network optimized by genetic algorithm was adopted to establish prediction model of grain moisture, then the prediction model is applied to the detection experiment of grain moisture content by NIR.Experimental results show that the prediction model used in this paper has more application value, because compared with BP neural network, the determination coefficient is improved by 1.356%, and the measurement accuracy of the instrument is 94.67%.