Accurately predicting soil carbon content for carbon sequestration projects is often difficult due to the amount of spatial variation in soil C content. As such, required sampling intensities can be high even for small areas. By quantifying spatial variability and spatial autocorrelation for different landscapes or land cover types, however, it is possible to adjust the required sampling intensities. For this study, the effects of three types of landcover—pine, hardwood, and pasture—on soil carbon spatial variability and autocorrelation were analyzed in the Piedmont of Georgia. We hypothesized that pasture plots would be the most homogeneous followed by managed pine and unmanaged hardwood. A cyclical sampling pattern with 64 sampling locations in the 0-7.5 cm soil layer per hectare under two plots of each landcover type was utilized. Semivariogram analyses were used to quantify the spatial autocorrelation of soil carbon concentrations and contents over lag distances of ~8 to 120 m. The major range of correlation (i.e., the maximum distance between points within which there is a degree of correlation) was greatest for percent soil carbon under the hardwood land cover for both plots (~63m and ~119m). The pine and pasture plots averaged major ranges of 35 and 41 m, respectively. Average carbon contents, based on point samples of bulk density and C concentration average over the hectares, were 29.9, 19.8, and 19.4 Mg/ha for the hardwood, pine, and pasture plots, respectively. Instead of averaging point estimates for each plot, it is also possible to krige based on measured values and the semivariance to create continuous surface estimates of soil carbon contents. Depending on the spatial variability of measurements, developing kriged estimates rather than averages may provide more reliable total soil carbon estimates for carbon trading or other offset projects.