Conventional soil maps are what soil survey experts turn out after field soil survey and interpretation of corresponding aerial photos, and often used as major data sources of information about spatial distribution of soils, which is essential to watershed management and eco-hydrology. With the development of geographic information technique, traditional soil survey methods are already far from efficient to meet the requirements of soil information services. As they used to be based on the experts’ empirical model of thinking, their products are often hard to express, exchange and store; the qualitative characteristics they described of a soil entity are often inconsistent with the characteristics of its spatial distribution, which tends to lead to low accuracy of the survey; and they are very costly and also limited to certain regions, which makes it hard to have information updated. Therefore, how to make full use of the existing historical resources and data is very important to retrieving efficiently soil maps higher in accuracy from the available information in Digital Soil Mapping (DSM). In this study, from the conventional soil maps and terrain data extracted were data of soil type and environment factors, based on which, a soil environment relationship model was established using the spatial data mining method, and finally, reliability and accuracy of the mapping was validated by field sampling. The Nieshui river basin in Huajiahe Town, Hongan County, Huanggang City, Hubei Province was selected for case study. The conventional soil maps of the study area plotted during the Second National Soil Survey were used to demonstrate processes of the research. The proposed method consists of five major steps. 1) Select seven environmental factors that were closely related to the process of pedogenesis and establish a geographic information system (GIS) database, which should contain a modified soil parent material map and data of terrain factors (elevation, slope, aspect, plan curvature, profile curvature and topographic wetness index) extracted from 10 m resolution Digital Elevation Model (DEM). 2) Extract 1410 typical sample data of soil types and environment factors by following the principle of frequency distribution, so as to reduce noises and abnormal data that would often occur in traditional soil mapping, because traditional soil mapping used to be done manually and contain some hard-to-reflect knowledge ( or noise) of the experts’ about proper relationship models. It is, therefore, essential to have the data properly pretreated. 3) Retrieve detailed expertise implied in the soil map product, using the spatial data mining techniques. Compared with the other algorithms, the decision tree algorithm is the most suitable one for extracting and expressing knowledge of the soil-environment model. So, the See5.0 decision tree algorithm is selected to perform spatial data mining and hence, obtain knowledge of soil and environment relationships. 4) Predict soil spatial distribution through inferring and mapping in Soil-Land Inference Model (SoLIM) based on the soil-environment knowledge and environment data obtained. SoLIM uses similarity degree as measurement parameter and fuzzy logic as basis to calculate similarity between soils. Within a given pixel, a number of corresponding soils have a variety of similarity degrees, which can be represented in fuzzy membership degree. Finally, the soil type represented by the highest fuzzy membership degree among the similarity vectors of a pixel is defined as the soil type of the pixel. A soil type distribution map can be obtained by hardening the fuzzy membership degree map. A large number of case studies have demonstrated that SoLIM is a more accurate than the traditional manual subjective method in soil mapping. 5) Verify accuracy of the proposed method through sampling at 270 field validation points using three sampling strategies: regular sampling, subjective sampling and transect sampling. Results show that the soil map obtained through fuzzy inference provides more detailed information about soil spatial distribution than its corresponding conventional soil map and is about 11% higher in accuracy and significantly higher in number of patches. It is therefore concluded that the proposed method which retrieves soil-environment relationships from a traditional soil map is more accurate than the conventional mapping method in judging and delineating and more convenient for use to update soil maps.