Rangelands make an important contribution to carbon dynamics of terrestrial ecosystems. We used a readily accessible interface (COMET VR) to a simulation model (CENTURY) to predict changes in soil carbon in response to management changes commonly associated with conservation programs. We also used a subroutine of the model to calculate an estimate of uncertainty of the model output based on the similarity between climate, soil, and management history inputs and those used previously to parameterize the model for common land use (cropland to perennial grassland) and management (stocking rate reductions and legume addition) changes to test the validity of the approach across the southwestern United States. The conversion of small grain cropland to perennial cover was simulated acceptably (< 20% uncertainty) by the model for soil, climate, and management history attributes representative of 32% of land area currently in small grain production, while the simulation of small grain cropland to perennial cover legumes was acceptable on 73% of current small grain production area. The model performed poorly on arid and semiarid rangelands for both management (reduced stocking) and restoration (legume addition) practices. Only 66% of land area currently used as rangeland had climate, soil, and management attributes that resulted in acceptable uncertainty. Based on our results, it will be difficult to credibly predict changes to soil carbon resulting from common land use and management practices, both at fine and coarse scales. To overcome these limitations, we propose an integrated system of spatially explicit direct measurement of soil carbon at locations with well-documented management histories and climatic records to better parameterize the model for rangeland applications. Further, because the drivers of soil carbon fluxes on rangelands are dominated by climate rather than management, the interface should be redesigned to simulate soil carbon changes based on ecological state rather than practice application.