Abstract: Soil water content is an important factor affecting the engineering properties of soil. It is of great engineering significance to measure soil water content. In this paper, the active heating optical fiber method is used to measure the soil water content through optical frequency domain reflection(OFDR)technology. The maximum temperature rise value method(ΔT_max) is used to calibrate the soil water content. The functional model between soil water content and the optical fiber temperature characteristic value is obtained. Based on the test results, the BP neural network prediction model of the soil fiber temperature characteristic value changing with soil water content, electric heating power, and heating time is established. The results show that the four soil water content function models were found to have better fitting performance under the conditions of higher electric heating power and longer heating time based on the experimental results. With the help of the soil water content function model, the predicted value of soil water content can be obtained through the temperature characteristics of fiber for soil; however, the prediction accuracy of the function model is limited. The neural network prediction model of soil water content established in this paper has high prediction accuracy, and the average relative error of soil water content prediction is only 0.18%.The results show that the neural network prediction model of soil water content has higher accuracy and better stability than the soil water content function model.