Near infrared reflectance (NIR) spectroscopy can produce reliable soil carbon measurements given proper method development. Unfortunately, attempts to transfer calibrations developed using one soil type to another, often suffer a loss of accuracy. This study documents change in carbon determination accuracy of NIR across systems, and how additional calibration data affects accuracy after transfer between narrow and broad geographic regions. To accomplish the goal, three sets of soil samples were used to calibrate and validate NIR carbon assessment models. These sets were taken from regions with differing agricultural systems, soil types, and carbon levels. Two of the regions were from two fields under a variety of different cropping systems in the Central Basin and Gulf Coastal Plain physiographic regions of Tennessee. The third set is from many different agricultural fields across the state. Four sets of calibrations were used. The three narrow regions were used to create individual NIR models and a fourth, broad model, was developed using a combination of all narrow regions. These models were then applied each of to the sample regions to measure loss of predictive quality. Results of the NIR model development and changes in accuracy caused by transference to differing regions will be presented.