Statistical relationships between site fertility type, proportion of deciduous trees, tree stem volume and multisource image data have been studied. The image data consisted of ERS-1 SAR images, Landsat TM image, NOAA AVHRR LAC image, airborne gamma radiation measurement data, digital soil map data, and geochemical data analyzed from soil samples. The 40 km by 30 km study area, centered at 61 degree(s)10' N, 25 degree(s)08' E, was in Southern Finnish Boreal forest. The ground truth data were boundaries of forest stands and the stand characteristic values for each stand. The analysis methods were linear regression analysis, discriminant analysis, and t-test. In regression analysis, the best model with one predictor variable, two predictor variables and so on were computed up to a model with ten predictor variables. At every step, all the potential predictor variables were tested. The best single image feature was the 1.5 micrometers channel 5 of Landsat TM. The best image variable type after Landsat TM data was the SAR image that had been acquired in March during snowing. Generally, winter SAR images were better than the summer image. SAR data produced a marginal utility of approximately 5 percentage units in estimation of tree stem volume. Use of several SAR images simultaneously increased the performance of estimation only little. On mineral soil lands, simple models gave results almost as good as more complicated models. On peatlands the performance of the models clearly increased when the number of predictor variables increased. The benefit of complicated models as compared to Landsat models only varied from 5 to 15 percentage units.