Soil organic matter (SOM) is an important soil component playing a critical role in various chemical and physical processes in soil, and meanwhile, may serve as a crucial index in assessing soil carbon stock, soil fertility and soil quality of an agro-eco-system. Therefore, to reveal the law of SOM spatio-temporal variability is a precondition for realizing sustainable utilization of soil resources. Then how to collect soil samples is the first issue that has to be dealt with in quantitative expression of spatial distribution of SOM content. And moreover, characterization of SOM spatial variability is closely related to size, density and distribution of soil sampling. Changes in the number and distribution of sampling sites may directly influence accuracy of the deduction and prediction of SOM spatial variability. On the other hand, generally speaking, usually the bigger, the number of samples to be collected, the more, the human resources, material and time to be consumed in sampling and lab analysis, but if the number of sampling sites is too small, it is very likely that some important spatial information of soil properties would be missed. Consequently, it is very meaningful to find out how the number of sampling sites affects characterization of SOM spatial variability and it is of some important practical significance to have a scientific designing of soil sample collection in an attempt to save soil sampling cost. On the basis of the 351 SOM valid data, a field experiment was laid out in Luhe County of Nanjing, Jiangsu Province to have 5 series of sampling designs, different in number of sampling sites, that is, 300, 250, 200, 140 and 100, set in line with the completely random sampling (CRS) and minimum sampling distance sampling (MSDS) methods. Based on deviations in variability deduction and spatial prediction of the 100 repeated samplings at each sampling series, effect of number of sampling sites on characterization of SOM spatial variability was explored. Results show that the nugget-to-sill ratios (C₀/Sill) of SOM derived from soil samples collected using two different sampling methods decreased with decreasing number of soil sampling sites, and it is more obvious with MSDS than with CRS, indicating that properly reducing the number of sampling sites may minimize the effect of sampling sites mismatching to SOM variation scale on variability deduction and hence improve reliability of characterization of SOM spatial variability. Comparison of SOM predictions using OK in deviation shows that using either sampling methods, RMSEs of spatial prediction, though fluctuating with decreasing number of sampling sites, were lower than that derived from all the sampling sites as a whole. The minimum RMSE was achieved when the number of sampling sites was reduced to 250 and the MSDS method was used, and about 6% lower than that of the prediction based on all the sampling sites as a whole. All the above findings imply that rationalizing the layout of sampling sites in number and density to match density of sampling sites to SOM variability scale is more important than simply increasing the number of sampling sites.