Tests of data processing approaches using Landsat data from a complex forested scene in California show the relative performance of ratioing, filtering and principal components approaches. Untransformed channels appear to perform best overall. Performances for untransformed, ratioed and the principal components of Landsat data channels are comparable for forest classes having complete cover. Filtered Landsat dta channels performed poorest overall, but perform best on select forest classes. Performance variability appears to be related to variations in background reflectance, surface illumination and spatial pattern by class.