Site quality estimation based on tree indicator species has many advantages over the conventional site index method, particularly in hardwood-dominated forests of the southern Appalachian Mountains. Not only are many hardwood species difficult to measure accurately for height and age, because of their diffuse branching form and wood structure, but also many of the underlying assumptions of site index are typically violated because the stands are usually uneven-aged, consist of mixed species, have been high-graded, or have latent crown damage from ice storms. We conducted a study to investigate how well species composition and selected environmental variables could distinguish among broad categories of forest site productivity. We used data from 210 plots in the Forest Inventory and Analysis Eastwide Database for the mountainous counties of western North Carolina, on which 27 common tree species occurred. Promising results were obtained using on-site tree species to estimate three broad levels of basal area productivity, which are probably adequate for extensive, practical forest management. In a multinomial logistic regression model of the three productivity levels, significant (P<0.05) variables included stand elevation and the presence or absence of six species: chestnut and scarlet oaks, hickory, red maple, serviceberry, and sourwood. Elevation was an important environmental variable because productivity declined as altitude increased. Classification accuracy was rather low, only 61 percent. However, our preliminary results are promising and indicate that additional work is warranted for (1) refinement of a suitable measure of basal area increment that adjusts for variable initial stocking levels and (2) better definition of low and high thresholds of site productivity. Adopting indictor species for site classification could prove particularly useful in growth and yield equations that use tree lists to drive the models becuse plot inventory data could be utilized also for site quality determination.