Relationship between soil organic carbon (SOC) and environmental variables has long been studied for estimating the SOC density at the unsampled locations. Considering non-stationary relationship at local scale, and by incorporating the regression residuals within the kriging, estimations can be improved. Therefore, the geographically weighted regression kriging (GWRK), a local spatial statistical approach was used in the present study to estimate the SOC across the Midwestern United States. Specific objectives of the study were to estimate the SOC density (kg C m^-2) and total SOC pool (Pg C) of the study area using the National Soil Survey Center database, consisting of 3,485 (calibration, n = 2788; validation, n = 697) georeferenced profiles. The normalized differential vegetation index (NDVI), temperature, land use, precipitation, elevation, and bedrock geology were used as predictors in the interpolation process. Results from this study support the conclusion that GWRK produced satisfactory predictions with lower root mean square error (5.60 kg m^-2), mean estimation error (0.01 kg m^-2) and mean absolute estimation error (4.30 kg m^-2), and higher R² (0.58) and goodness-of-prediction statistic (G = 0.59) values. The global regression explained only 28% of the spatial variations, whereas, local regression (GWR) explained 45%. Croplands of the region store 16.8 Pg C followed by shrubs (5.85 Pg) and forests (4.45 Pg). Total SOC pool for the Midwestern region ranges from 31.5 to 31.6 Pg. This study illustrates that GWRK approach explicitly addresses the spatial dependency and spatial non-stationarity issues for interpolating SOC density across the regional scale.