There has been a great concern about how to accurately predict soil properties using the limited soil samples. At present, the approaches of Kriging interpolation coupled with auxiliary variables has been widely used. However, little information on improving prediction accuracy of soil organic matter (SOM) and total nitrogen (STN) with the aid of land use patterns as auxiliary variables is available. In this study, 254 soil samples were collected in Yujiang County of the hilly red soil region, China, two approaches: ordinary kriging (OK) and kriging combined with land use patterns information (KLU) were used to predict SOM and STN spatial distribution pattern, and 102 samples were validated to compare the prediction accuracy of these two approaches. The results showed that the correlation coefficients between measured and predicted SOM and STN values using KLU approach (rSOM=0.786, rSTN=0.803) were both great larger than those using OK approach (rSOM′=0.224, rSTN′=0.307). As for 102 validated samples, the root mean square error (RMSE) of SOM and STN using OK approach were 12.48 g·kg-1 and 0.64 g·kg-1, while those using KLU approach were 6.86 g·kg-1 and 0.37 g·kg-1, which were 55% and 58% of the former only. In terms of KLU approach, RMSE of drylands has the widest lowering range, and that of forestlands has the smallest lowering range. It is indicated that KLU approach is an efficient and practical prediction approach in the hilly red soil region, China.