Based on dense soil samples collected from the middle-area of Yujiang County to get 4 aggregation grades of samples distribution by resample which included 5 repetitions in each one, the VMRs which represent the mean variance of samples in quadrat analysis were 0.13, 0.83, 1.49 and 2.16, respectively. Ordinary kriging (OK) and kriging combined with land use pattern information (KLU) were used to predict soil total nitrogen (STN) spatial distribution pattern, and 40 samples were validated to compare the prediction accuracy of these four aggregation grades, and to reveal the aggregation grades of samples distribution impact on prediction accuracy of STN. The results showed that the correlation coefficients r between measured and predicted STN contents from OK and KLU were decreased with increasing aggregation, and the r values reduced from 0.400 to 0.142 for OK and from 0.718 to 0.542 for KLU, respectively. The mean absolute errors (MAEs) and the root mean square errors (RMSEs) of STN from OK and KLU increased with increasing aggregation, and MAEs increased from 0.49 to 0.61 for OK and from 0.33 to 0.44 for KLU, and RMSEs increased from 0.56 to 0.65 for OK and from 0.39 to 0.47 for KLU, respectively, indicating that on the premise of the same number of samples, the lower aggregation that the samples distribution more uniform, the higher prediction accuracy by kriging. It shows that using the regular grid sampling is more conducive to spatial estimation when spatial prediction for STN, and the spatial aggregation of samples also have different impacts on the prediction accuracy with different kriging, and the impacts on KLU is greater than OK.