In order to study the spatial heterogeneity of soil organic carbon and the different sampling scales have an impact on the prediction accuracy of the spatial distribution of soil organic carbon in Karst area. Using grid sampling method, the soil organic carbon content under the sampling of 150 mx150m mesh is taken as the original data, Select a subset of samples at 5 grid scales (300mx300m, 450mx450m, 600mx600m, 750mx750m and 900mx900m). The spatial variation and layout are analyzed and predicted by the ordinary Kriging interpolation method. As shown by the results, the sampling scale significantly affected the prediction accuracy of spatial distribution of soil organic carbon. On the scale of Karst watershed, soil organic carbon research was taken as the purpose of investigation. In soil sampling, 450mx450m sampling scale is a suitable sampling method. When the soil organic carbon is studied on the Karst watershed scale, if the Cochran formula to determine an optimal sampling quantity is used blindly, an insufficient number of sampling points could be determined, which would negatively affect the accuracy of the soil organic carbon spatial prediction. The data derived from low-density sampling are used to fit the trend, whereas the data from high density sampling are used to fit the variation function. Meanwhile, an auxiliary variable for the soil organic carbon is introduced for spatial prediction, which may improve the prediction results. Spatial variability of soil organic carbon in Karst should be taken into full account. It is more important to set the distance of soil sampling points and the number of sample spots than to increase the number of sampling points.