AbstractSpectroscopy is a powerful means of increasing the availability of soil data necessary for understanding carbon cycling in a changing world. Here, we develop a calibration transfer methodology to appropriately apply an existing mid infrared (MIR) spectral library with analyte data on the distribution of soil organic carbon (SOC) into particulate (POC), mineral-associated (MAOC), and pyrogenic (PyC) forms to nearly 8000 soil samples collected in the Great Plains ecoregion of the United States. We then use this SOC fraction database in combination with a machine learning-based predictive soil mapping approach to explore the controls on the distribution of fractions through soil profiles and across the region. The relative abundance of each fraction had unique depth distribution profiles with POC fraction dropping exponentially with depth, the MAOC fraction having a broad distribution with a maxima at 35–50 cm, and the PyC fraction showed a slight subsurface maxima (10–20 cm) and then a steady decline with increasing depth. Within the Great Plains ecoregion, clay content was a strong control on the total amount and relative proportion of each fraction in both the surface and subsoil horizons. Sandy soils and soils in cool semi-arid regions contained significantly more POC relative to the MAOC and PyC fractions. Cultivated soils had significantly less SOC than grassland soils with losses following a predictable pattern: POC > MAOC ≫ PyC. This SOC fraction database and resulting maps can now form the basis for improved representation of SOC dynamics in biogeochemical models.