The study involved the delineation and analysis of forest ecosystems in the Cyrus H. McCormick Experimental Forest, Upper Michigan, U.S.A. There were three general objectives: to develop a local ecosystem classification using a multifactor approach, to determine the most efficient method of predicting ecosystem membership, and to study the ability of ground flora to predict edaphic factors. Special attention was given the role of physiography and soils in ecosystem classification and the functional relations between plant and soil. The study area was stratified into biologically equivalent ecosystems using a special reconnaissance field technique. Sixty-six plots were sampled using stratified random sampling. Detailed observations were recorded on physiography, soils, and vegetation. Standard soil laboratory analyses included particle-size distribution, organic matter, pH, and macronutrients. Statistical analyses were used to compare ecosystems and test the validity of the field classification. The analyses included multiple linear regression, analysis of variance, principal component analysis, and numerical clustering methods. Multivariate discriminant analysis and canonical ordination were used to predict ecosystem membership. Unbiased probabilities of misclassification based on discriminant analysis were calculated using the jackknife method. Empirical distribution functions and rank order correlation were used to study the distribution of ground flora species over environmental gradients. Eleven dryland forested ecosystems were identified using a multifactor approach. Each ecosystem occupied a characteristic topographic position within the landscape. Individual ecosystems significantly differed in many of their biophysical properties such as slope, aspect, soil texture, soil drainage and soil fertility. The ecosystems were characterized by distinctive potential overstory and ground flora compositions. Deep, well drained soils with more than 10% silt and clay had overstories strongly dominated by Acer saccharum Marsh. Extremely xeric sites, either very well-sorted sands or rocky soils, supported overstories dominated by Pinus banksiana Lamb., Pinus strobus L., Quercus rubra L. and Acer rubrum L. Imperfectly drained sites had a diverse overstory vegetation. Differences in forest composition and environmental factors showed close correspondence over the late successional forests of the study area. Placing the emphasis of ecosystem classification on the biophysical structuring of the local landscape greatly facilitated the understanding of the functional relations among plant and environment. The most reliable and efficient method of local ecosystem classification was the use of a combination of physiography, soils and vegetation. The combined approach classified ecosystems 18% more efficiently than the use of vegetation alone and 5% more efficiently than the use of physiography and soils alone. Relations among ecosystems using ordination techniques depend fundamentally on the data used to produce the ordination. Ordinations differed when comparing physiography and soils data with vegetal data. A combination of both produced the best ordination. Certain ground flora species, combined into ecological species groups, were found to be accurate and effective indicators of soil moisture, soil texture, soil pH and soil total nitrogen. It was possible to make probability statements about specific edaphic conditions merely by noting the presence or absence of selected ecological species groups.