The purpose of this study was to analyze the relationships between soil attributes and environmental covariates in a tropical hillslope environment on a regional scale to estimate spatial distribution of soil attributes and identify statistical and geostatistical techniques that could represent the variation of the soil attributes. The studywas performed in Bom Jardim County, Brazil, and covered an area of 390 km2 with a soil database of 208 sample points distributed in six depth layers (0.53 pts/km2). The study used 18 environmental covariates derived from DEM and satellite imagery. The models evaluated were linear regression, regression trees and ordinary and regression kriging. An exploratory analysis showed that DEM, NDVI, MRVBF, MSP, b3/b2, b5/b7, SPI, SWI, SLOPE and ASPECT were correlated with soil properties. The models performance had a mean crossvalidation r2 of 0.13. The best results were achieved with kriging models, with a crossvalidation r2 of 0.19. A comparison between multiple linear regression and regression trees showed that the tree model yielded the best results. The sample density alone could not explain the results, but an interaction between DEM accuracy, sample density, covariates and geological conditions was suitable as an explanatory factor. Studies of tropical hillslope digital soil mapping on regional scales need to be more exhaustively focused to develop this research area.2014