Prediction accuracy for projected basal area and trees per acre was assessed for the growth and yield model of the Forest Vegetation Simulator Southern Variant (FVS-Sn). Data for comparison with FVS-Sn predictions were compiled from a collection of n 1,780 permanent inventory plots from mixed-species upland hardwood forests in the Southern Appalachian Mountains. Over a 5-year projection interval, baseline FVS-Sn predictions fell within 15% of observed values in over 88% of the test plots. Several modifications to FVS-Sn were pursued, including a refitting of the background mortality equation by logistic regression. Following the modifications, FVS-Sn accuracy statistics increased to 91 and 94% for basal area and trees per acre, respectively. In plots with high initial stand densities, notable gains in accuracy were achieved by relaxing thresholds that activated a density-dependent mortality algorithm in FVS-Sn. Detailed accuracy results for forest types of the region were generated. Twenty-five-year projection results show size-density trajectories consistent with the concept of maximum stand density index.