The prediction model for liquefaction potential of site based on artificial neural networks is presented. The spatial data structural characteristics of liquefaction potential of site can be mapped by different spread values. The optimal spread value can be determined by means of cross-validation technique of geostatistics. Study results indicate that the generalized regression neural networks can map the structural characteristics of liquefaction potential under the condition of the optimal spread. The result obtained is consistent with that of Kriging estimator. The model can be applied to the prediction of two-dimensional spatial data and the decision system based on GIS.