Using the environmental information gathered and results of the numerical classification obtained by fuzzy c-means algorithm, regression models of soil taxonomic distance-environmental covariates were developed after spatial autocorrelation of taxonomic distance, to predict the taxonomic distances between soils at any spatial positions and the central profiles established in the study area. Based on the above prediction, a digital soil map with grid resolution of 25 m was produced after defuzzification and fragment-incorporation. It was clearly shown in the digital soil map that Typic Endorusti-ustic Cambosols was soil type with the biggest distribution percentage, covering about 36% of the total study area, while Parasalic Siltigi-ustic Cambosols, the soil type with the second biggest coverage, occupying around 24% of the land surface. By analysis on spatial variability of both determinable trend and undeterminable residual involved in soil taxonomic distance, the influences of soil-forming environment and stochastic factors on soil spatial variation were discussed. It was concluded that, compared to predictive map originated from a stochastic model, the digital map produced by the employment of soil taxonomic distance-environmental covariate regression model displayed a similar pattern of spatial soil distribution. However, much more information contained and reasonable delineation borders in the digital map was indicative of its capacity to demonstrate continuity and transition of soil distribution, and to explain spatial soil variation induced by micro-environmental factors and human activities.