Log prices can vary significantly by grade: grade 1 logs are often sev
eral times the price per unit of grade 3 logs. Because tree grading ru
les derive from log grading rules, a model that predicts tree grades b
ased on tree and stand-level variables might be useful for predicting
stand values. The model could then assist in the modeling of timber su
pply and in economic optimization. Grade models are estimated for ten
species groups found in the southern Appalachians, using data from sev
eral thousand trees and permanent plots in the USDA Forest Service's F
orest Inventory and Analysis (FIA) database. The models correctly pred
icted grades of a majority of trees in both a test and a validation da
ta set, and predictions of grade proportions across a sample of the po
pulation were usually within three percentage points of actual grade p
roportions. But success of models varied across species and diameter g
roups. Considering several measures of modeling success, the most accu
rate models were those predicting tree grades for softwoods and larger
hardwoods.