Abstract: Neural Kriging(NK) model was established by BP model of artificial neural network, which possesses the similarity of operational objectives to ordinary Kriging(OK) and Conditional Simulation(CS). The NK model was applied to study the space variability of water-salt distribution during soil freezing and thawing periods—the initial freezing period, the maximum freezing depth period and thawing period in the cropland and non-cropland by simulating and testing sampling points and estimating unknown points. Comparing simulation, test and estimation results of NK model with that of OK model and conditional simulation and comparing semi-variogram of NK model with that of sampling value, OK estimated value and CS value. Results show that the NK method is better than OK method in eliminating moving-average effects. Furthermore, the NK method has itself particular advantages that do not require estimation of covariance function and semi-variogram treatment. At the same time, it has reasonably accurate estimation of prediction. So this method has more flexible adaptability for unique value and extend from one-dimensional to three-dimensional space than OK and CS method. And this method is a complement method for application of traditional space variability research. At the same time, it will broaden applied fields of artificial neural network(ANN) theory and has advantages of discipline interfusion.