Accurate prediction of organic matter distribution in soil is critical to sustainable soil management. Although correlated factors have been used to improve the accuracy of soil organic matter prediction, very few studies have considered the nonlinear relationships between these correlated factors and soil organic matter. The objective of this study was to use the clustering of self-organizing maps to describe the nonlinear relationships between soil organic matter and correlated factors and then examine whether ordinary kriging combined with the clustering of self-organizing maps (KCSOM) can improve prediction accuracy. The performance of the model in the Pinggu district of Beijing was compared with four interpolators: ordinary kriging, ordinary kriging combined with soil texture, ordinary kriging combined with soil type map delineation, and ordinary kriging combined with land use. Results showed that KCSOM accounted for the nonlinear relationships between soil organic matter and the correlated factors and was the only technique that effectively avoided underestimation of the higher values and overestimation of the lower values of the interpolation surface. Moreover, the spatial variation of soil organic matter for different clusters of an entire map was more accurate than spatial variation generated by ordinary kriging. The mean error, root mean squared error, and relative improvement for KCSOM were 0.004, 2.01, and 30.92%, respectively. The estimation imprecision of KCSOM was decreased by 77.04%. These results indicate that prediction accuracy was greater with KCSOM than with any of the other methods and that the proposed technique can serve as an effective method for prediction of soil organic matter.