Accurate estimates of the spatial variability of soil organic matter (SOM) are necessary to properly evaluate climatic chagne, soil carbon sequestration potential and soil fertility. In plains and gently undulating terrains, soil spatial variability is not closely related to relief, and thus digital soil mapping (DSM) methods based on soil-landscape relationships often fail in these areas. Therefore, different predictors or methods are needed for DSM in plains. In provincial regional scale, climatic factors influence spatial distribution of soil properties. For this research, Jiangsu Province was selected as example and mean annual temperature (MAT), mean annual precipitation (MAP), physical clay content, and soil pH were selected for SOM spatial modeling using geographically weighted regression (GWR). The SOM content in the surface layer (0－20cm) of 1 519 typical soil profiles of the Second National Soil Survey in Jiangsu Province were collected. 1 217 samples were selected as the modeling set and 302 were the validation set. Fristly, 100% (1 217), 80% (973), 60% (730), 40% (486), and 20% (243) samples were randomly selected from the modeling set, and global and local spatial autocorrelation of SOM content were analyzed at different spatial scales using spatial statistics tools in ArcGIS. Secondly, comparison of the accuracy between GWR model and the global regression model under the different sampling size was conducted. Akaike information criterion (AIC), residual sum of squares (RSS) and adjustment determination coefficient (R2adj) were used modeling comparison. Thirdly, the optimal model was selected for mapping SOM spatial prediction. Independent validation was used for model evaluation, using four indices: mean error (ME), mean absolute error (MAE) and root mean of squared error (RMSE), and determination coefficient (R2). Results show that: 1) There was a significant spatial autocorrelation of SOM content in Jiangsu Province at different spatial scales. The clustering pattern of global and local spatial autocorrelation of modeling set with different sampling size were similar. The global Moran's I ranged from 0.25 to 0.61 (P<0.001). The spatial distribution of SOM content was mainly characterized by spatial clustering pattern. The "high-high" clustering areas were mainly distributed in the central and south of Jiangsu, and the "low-low" clustering areas were mainly distributed in the north of Jiangsu. 2) The modeling results of GWR were better than the global regression modeling, and the residuals had no spatial autocorrelation at different spatial scales. The R2adj of GWR in different modeling sets was increased by 0.15 to 0.20 compared with the global model. The AIC and RSS were significantly lower than the global model, which were decreased by 56.08 to 360.19 and 17.40 to 76.67 respectively. There were slight difference between GWR models with different sampling size. 3) The number of modeling samples (except for the number of modeling samples was 243) had little effect on the accuracy of prediction and mapping results of GWR, the RMSE was between 5.56 and 5.75 g/kg, MAE was between 3.87 and 4.05 g/kg and R2 was between 0.48 and 0.52. The results were all better than the validation result of Ordinary Kriging using all modeling sampling points. This study can provide reference for SOM modeling and mapping in large and low relief areas with sparse samples.