Based on field investigations conducted in the 1980s of Wenquan District, Xinghai County, Qinhai Province, a large tract of permafrost on the east edge of the vast Qinghai-Tibetan Plateau (QTP), new approaches to soil mapping were explored. Based on the theory of soil forming factors and the theory for soil-landscape modeling and the screening of potential variables in soil classification, a soil-landscape model was built up using SVM coupled with various combinations of the variables, and applied to predictive soil classification of the studied area. To better verify effectiveness of the new approach, a 5-fold cross validation method was used. By comparing the outcome of the cross validation of the various variable combinations with the simulated distribution map, a set of combination of environmental variables was defined suitable for soil classification of typical permafrost area. Thus fewer samples are needed to better predict spatial distribution of the types of soils in the area.