Three fundamental questions structure the research: 1) How skillful are dynamically downscaled models in simulating minimum and maximum temperature and mean precipitation in ahistorical reference period (1970-1999) for the Southeast United States? 2) What are the magnitude of biases for each NARCCAP member (and variable) and what is the potential source of the bias? 3) Does downscaling improve projections at local scales? In other words, is value added in downscaling? Analysis was performed on the states encompassing Alabama, Mississippi, and Tennessee (west sub-region), and Georgia, North Carolina, and South Carolina (east sub-region). Skill was determined using three methods: 1) Computing the overlap in probability density functions (PDF) between observations and models, 2) computing an index of agreement between models and observations, and 3) computing the root mean squared error (RMSE) between observations and models. Most models illustrated high skill for temperature. The outlier models included two RCMs run with the GFDL as their lateral boundary conditions; as these models suffered from a cold maximum temperature bias, attributed to erroneously high soil moisture. Precipitation skill using the PDF and index of agreement methodologies showed high skill for all models; however, the RMSE-based skill showed some models' skill suffered from over estimating the frequency of extreme precipitation events.