Soil organic matter (SOM) is a well known indicator of soil quality and has the potential to counter balance the increasing atmospheric concentration of greenhouse gases. However, there is a nationwide need for site and condition specific soil organic carbon (SOC) data to increase the reliability of estimating amounts of C that can be practically stored or emitted by soils by changing land use/cover. As part of the effort to create a scientifically- and statistically-based U.S. soil C stock inventory, the present study had the objectives to i) develop and validate visible-near infrared (VNIR) models using spectral library at soil survey laboratory (SSL) and ii) predict SOC values for samples collected in a watershed in western Louisiana to evaluate the effects of soil type and land use/cover on SOC stocks. The spectral library of the Major Land Resource Area Regional Office (MO) where the watershed is located had 566 samples. VNIR models relating soil spectra and SOC contents were fitted for both the whole data and subsets data stratified by horizon designation (A horizon and subsurface horizons). Partial least square regression was used for fitting the models. Data preprocessing included Savitsky-Golay first derivative with eleven and twenty one band window, multiplicative scatter correction, standard normal variate, and unit area normalization. Across two validation rounds with two independent validation sets, the average root mean square error validation (RMSEv), residual prediction deviation (RPD), and coefficient of determination (R²) for the selected models ranged 1.07–6.91 g kg^-1, 1.56–1.84, and 69–76%, respectively. Preliminary predictions of the watershed samples have indicated a more accurate SOC content determination by using subset data models stratified by horizon designation than by using whole data models.