In order to establish fine soil water forecast model in middle area of Shandong province, the daily dataset of the farmland automatic soil moisture station were used to analyze the annual and monthly soil moisture variations from 2010 to 2013. Combined with the dataset of automatic meteorological stations data nearby, based on the soil water balance principle and field evapotranspiration model, the stepwise regression and curvilinear regression methods were used to establish the empirical forecast model of daily and seventh day's soil moisture declining in condition of continuous nonprecipitation from April to June in irrigated plain and mountainous area. Results show that the annual variation trend of soil moisture storage from 0 to 100 cm depth was consistent with the depth from 0 to 50 cm in the middle area of Shandong province, whose maximal point appeared in August, the minimal point in June, and the largest monthly decreasing amplitude appeared in March to June, so drought was prone to happen in those months. The back substitution and forecast results of the models show that the average relative error of the back substitution was 0.07%, and the absolute errors and relative errors of the seventh day's soil moisture storage by using the seven days' model and one day's model from 0 to 50 cm depth were -0.15 mm and -2.17 mm, -0.07% and -1.56%, respectively. The models are characterized by high accuracy, strong practicability and theoretical basis, which could provide support for soil moisture monitoring and fine forecast in the middle area of Shandong province.