It is a challenged task to examine the spatial variation of soil with sparse measured data. In this study, the general regression neural network (GRNN) method was presented to predict the 3D geologic soil profile of a region based on the data of seven boreholes. A probability vector was introduced to represent the soil type at the associated location. By ensuring the difference between the predicted soil type and the measured one less than 1% at these measured boreholes, the smoothing parameter σ in the GRNN method was determined. With the selected smoothing parameter, the regression model was developed using the borehole data and a 3D geologic model was established to present the spatial distribution of soil type. It indicates that the GRNN method can realize a simple and intuitive geologic modeling using only the spatial coordinates.