The main environmental factors of the influence of soil temperature are air temperature, total solar radiation, effective photosynthetic radiation, wind speed and relative humidity, while the farmland water and submerging duration are the main factors affecting the change of soil temperature, and it exist complex non linear relationship between them. The aim of this study was constructed the soil temperature prediction model on farmland water level and environmental factors and obtained paddy soil temperature predictive value by using the data of existing water level and environmental factors. For the flooding treatment, the minimum and maximum soil temperature appeared at 7:00 and 18:00, for the drought treatment the minimum and maximum soil temperature respectively appeared at 6:00 and 14:00. This study trained artificial neural network based on back propagation algorithm (BP-ANN) through the existing data to predict the four characteristics the soil temperature of different water level control so as to obtain the soil temperature amplitude. Results showed that there was certain deviation between the predictive value and the actual value at the four characteristic moments, relative error were 1.19%, 1.34%, 2.09% and 1.07%, the predicted outcome was satisfactory. It is significant for guiding the rice irrigation and the production of practice facilitated.