[Objective] Soil organic carbon (SOC) varies sharply with time and space, as it is always subject to influences of various soil-forming factors, natural environmental factors and human activities. So to determine how an appropriate number of sampling sites could affect accuracy of the prediction of SOC in different time periods is the basis for formulating a scientific strategy for high-efficient soil sampling.[Method] In this study, a tract of upland (3.93×10⁴ km²) in North Jiangsu was delineated and selected as a case area, and the Ordinary Kriging interpolation method commonly used in soil science was adopted in analyzing influences of numbers of sampling sites on prediction and mapping of SOC in different time periods. The study was designed to have 20 treatments, which were set in accordance with the principle of 5% decrease in number.[Result] Results show that with the number of soil sampling sites decreasing from 100% to 5%, correlation coefficient (r) between the predicted value and the measured value of SOC in 1980 and in 2008 varied in the range of 0.15-0.56 and 0.24-0.63 and root mean square error in the range of 2.09-2.63 and 2.11-2.62 g·kg^-1, respectively. As in 1980, the SOC in the studied region varied quite slightly in spatial autocorrelation and quite drastically and locally, its prediction improved slowly and unsteadily in accuracy, and around 563 samples were needed to make the prediction relatively reliable. However, in 2008, the SOC in the region varied quite sharply in spatial autocorrelation, but mildly locally, and hence its prediction was very sensitive in accuracy to variation of the number of sampling sites. So 526 soil sampling sites were enough to ensure stable prediction accuracy. Standard root mean square errors of the 20 treatments in terms of number of sampling sites varied in the range of 0.34-0.43 and 0.20-0.25 g·kg^-1, in 1980 and in 2008, respectively, and spatial prediction was higher in 2008 than in 1980 in accuracy when the numbers of soil sampling sites were the same.[Conclusion] Results of this study indicate that the optimal number of soil sampling sites and their prediction accuracy in the same area are not fixed, but determined in the light of spatial variability of soil attributes, distribution and spatial layout of the sampling sites in each time period. As environment, climate and farmland management practices all vary with time period, SOC content does too in spatial structure and layout, which will greatly affect the optimal number of sampling sites relative to time period.