Dimensional analysis techniques were used in the harvest of 31 bigtooth aspen (Populus grandidentata Michx.) from three stands representative of the range of soil quality among aspen sites in northern lower Michigan. Mean tree age at the sites ranged from 52-60 years. Biomass and annual production of each aboveground component were individually measured. Original methods were developed for estimation of stembark biomass and the twig weight of branches. Regression equations were derived to predict component biomass and annual production from tree dbh. Evaluation by analysis of covariance indicated significant differences in regression models among the sites. Differences of stem component regressions were due to corresponding differences in tree height to dbh relationships. Variations among regressions for crown components (live branch wood with bark, leaves, twigs) were probably a result of differences in light attenuation and competitive interaction effects among sites. Regressions derived from individual site harvest data, plus data from sites with statistically equivalent component weight to dbh relations, were used to determine aboveground biomass and annual production of bigtooth aspen. In 1975, total aboveground biomass was 17156, 12877, and 3853 g/m2 at the good, intermediate, and poor sites, respectively. Corresponding aboveground annual production values were 1104, 726, and 293 g/m2. Component percentages of total biomass generally were similar among sites, except for leaves. Variations in production percentages showed a production per unit leaf weight gradient parallel to the site quality gradient. Biomass and production values were compared to those reported for other Populus sp. forests. The errors inherent in the substitution of regressions derived from data from other sites were examined. Total biomass estimates ranged from -27 to +40 percent of accepted values. Errors for individual components ranged from -33 to +51 percent. Use of regressions derived from the combined data from all sites also produced significant but less extreme errors in several cases.