Abstract: Soil thermal conductivity is of great significance to the development and utilization of shallow geothermal energy and underground space,and the design of reservoir energy storage and nuclear waste disposal. On the basis of previous studies,volume fraction of solid,liquid and gas in soil of the same type in the same region were selected as independent variables and thermal conductivity as dependent variables,and SPSS software RBF neural network was used for nonlinear prediction. The data of high pressure chamber bentonite,red clay and silty sand collected in the literature were calculated and analyzed to verify the superiority of this method. Then the data of the geological exploration project of shallow geothermal energy resources in Beijing plain area and the shallow geothermal energy resource survey and evaluation project in Xiaoshan district of Hangzhou city were calculated and predicted,all the average relative errors of the predicted results to the measured values are less than 5%.