Rapid development of geographic information systems has facilitated modeling soil spatial variability and pedological processes using landform attributes calculated from digital elevation models (DEM). Although correlations between soil properties and the topographic variables are strongly determined by the grid size of the DEM, few studies of soil-landscape analysis have examined such an influence under multiple spatial scales. This study shows how the predictability of three adaptive techniques (generalized linear model, GLM; artificial neural network, ANN; regression tree, RT) varied depending on the selection of spatial resolutions for DEM-based terrain analysis. We collected 193 soil samples from the surface of Sindu coastal dunefield at western Korea and analyzed eleven physical and chemical soil properties. Based on a principal component analysis, four soil attributes were selected to test for environmental correlation, assuming that they reflect dominant pedogeomorphological processes on the dune. Predictors included vegetation types, distance from the seashore, and terrain parameters extracted from various grid sizes of DEM (5, 10, 20, 30, 40 m). The three models were most successful in predicting the spatial distribution of soil properties that represented the effects of nutrient input from marine sources by aeolian processes and topographic relief, an indirect measure of available soil moisture. ANN, in general, showed the highest prediction accuracy, while its predictability significantly varied as the grid size of DEM changed. GLM and RT had a relatively similar performance. However, GLM was the least sensitive, and thus the most stable model regardless of the spatial resolution. We conclude that there is no absolutely best model and spatial resolution for predicting soil spatial variability. Future pedogeomorphological modelers therefore need to be careful when determining optimal models and spatial scales of DEM in order to take into account the unique spatial extent and behavior of individual soil attributes.