The study area consists of the Finnish part of a Landsat 5 TM image from 1990. Three independent field samples were measured during 1991–93 in the study area. The first sample was used to compile training areas for supervised maximum likelihood classification of the image. Classification accuracy was studied in the second sample. The spectral separability of the forest strata usable in practical forestry was poor. The extent of the damage area was estimated by the principle of stratified sampling. The estimate included considerable bias because the field sample had not been objectively selected from the image classes. The third field sample was measured as part of the National Forest Inventory of Finland. It is wholly objective, and about ten times larger than the two earlier field samples. The poor spectral separability of the forest strata was confirmed by the NFI sample. However, this sample could be used in stratified sampling with little or no bias in the estimation of the damage area estimate. 14 different damage types were separated according to specific damaging agent. A thematic map was produced which presents the spatial distribution of two damage-rich image classes. The study area comprises 18 300 sq.km, of which 38% were damaged. At first sight it would appear that the proportion of damaged forest has tripled in ten years. However, this is not the case because now special attention was paid to forest health in the field work. Despite this, it is possible that some damage caused by unfavourable climatic phenomena in the ’80s was still perceptible in 1993. No damage caused directly by air pollution has yet been verified in the study area.