Soil organic carbon (SOC) in Beijing was taken as target variable and four different sampling densities were designed to investigate the structural changes of the variogram and uncertainty of spatial prediction with the study scale changes. The results showed that the mass ratio of SOC was macroscopically related to terrain factor and low sampling density data were the most optimal for use in fitting the trend values. As sampling density increasing, the variogram distribution of SOC mass ratio and its residuals flattened out gradually. The random variation was growing strongly, and the structural variation and uncertainty of spatial prediction decreased gradually. In addition, the range of variogram might also affect the uncertainty of spatial prediction. Increasing sampling density and regression Kriging method aided by terrain factors can improve the prediction accuracy of mass ratio of SOC. Therefore, soil monitoring and management introducing auxiliary variable can cut the number of sampling points to some extent without reducing prediction accuracy.