soil-site relations for trembling aspen in northwestern ontario were studied using 98 site plots representing a wide range of glacial landforms, soil conditions, and site quality. site index was related to features of soil and topography using multivariate statistical analyses including principal component analysis, multiple regression analysis and cluster analysis. site index ( si = height of dominant and codominant trembling aspen trees at 50 years breast-height age) was used as the dependent variable; 45 soil and topographic values were used as independent variables. principal component analysis combined with correlation coefficients was used to select 10 variables from 22 variables that were closely associated with site index; these 10 variables are not closely correlated with each other. preliminary regressions indicated that the soil-site relationships were much better expressed when plots were stratified into three landform types as opposed to a single regression combining all plots. final regression equations were computed describing soil-site relationships on soils developed from glaciofluvial, morainal, and lacustrinal landforms. the final regression equation for the glaciofluvial soils included depth to root restricting layer and drainage class as site index predictor variables. the final regression equation for the morainal soils included silt plus clay content of the a horizon, coarse fragment content of the c horizon, and depth to root restricting layer. the final regression equation for the lacustrine soils included the clay content of the c horizon, and depth to mottles. the 98 plots were clustered into six groups representing different soil conditions. the fec soils s3, s4, s5 were the best sites; the ss8, ss7, s7, s8, ss5, ss4 were the worst sites for aspen. however, cluster results were not significant due to a wide range of site indices within each of the six defined groups. large standard deviations and standard errors of the mean exist in most groups. thus the use of these groups is not recommended for estimating aspen site index.