- Concentration of carbon dioxide (CO₂) in the atmosphere has increased over the past 150 years. Because CO₂ is one of a number of radiatively active gases, there is concern that global temperatures will rise and climatic conditions will change. Recent research indicates northern hemisphere forests may currently be accumulating carbon (C) from the atmosphere. Live trees hold a large proportion of forest C, however, live-tree C can only be measured indirectly and therefore estimates of live-tree C are subject to numerous uncertainties. The objectives of this research were to estimate how live-tree C stores changed in the Pacific Northwest (Oregon and Washington west of the Cascade crest) between 1963-91 and to assess the factors introducing uncertainty into the estimate of live-tree C storage. The first objective was accomplished by using data from the Forest Service Forest Inventory and Analysis Program (FIA), combined with western Oregon and western Washington annual timber harvest data. The study produced live-tree C estimates for all timberland by land-ownership group. Between 1963-91, C on all timberland in the Pacific Northwest decreased from 1636 to 1392 Tg, or by 15% of the 1963 total. National forest, other public (other federal, state, and local government), forest industry, and miscellaneous private land lost 15, 5 (non-significant), 24, and 18% of their 1963 total timberland live-tree C by 1991, respectively. All landowners except industry experienced significant declines in total timberland area. C density (live-tree C per area) on all timberland dropped by 13% on national forests and by 30% on forest industry, but rose by 1% (non-significant) on other public and 26% on miscellaneous private land. For the Pacific Northwest as a whole, C density on all timberland decreased by 8% over the 28-year study period. C density declined most dramatically between 1963 and 1974. Since 1974, increasing C density on other public and miscellaneous private land balanced declining C density on national forest and forest industry land, resulting a C density ranging between 135-136 Mg C ha⁻¹ on all timberland. The live-tree C estimate is subject to uncertainty arising from sampling, regression, measurement, and model error. We created and implemented a method for assessing uncertainty arising from model error. Volume equations, densities, biomass equations, and C:biomass ratios were compiled for the five major tree species in northwest Oregon: Picea sitchensis, Pseudotsuga menziesii, Tsuga heterophylla, Acer macrophyllum, and Alnus rubra. Volume equations were transformed into biomass by multiplying predicted volume with a range of species-specific measured densities. Biomass derived from volume equations multiplied by densities or from biomass equations was converted to C using a range of C:biomass ratios. For each tree component, species, and diameter at breast height, the maximum and minimum C predicted by equations was captured and stored as lookup tables. Component lookup tables were summed to create estimates of tree total C under three assumptions about within-dbh class correlation between components: perfect positive, zero, or perfect negative correlation. Application of lookup table bounds to individual tree data from the FIA program produced estimates of minimum and maximum C for the five target species in northwest Oregon. The above methods resulted in a base-case live-tree C estimate for northwest Oregon ranging from 28-210 Tg C (±76% uncertainty) assuming perfect positive correlation, and 67-1 54 Tg C (±40% uncertainty) for perfect negative correlation. When height variation was incorporated, C storage uncertainty rose to ±91% for positive and ±51% for negative correlation. A gain in precision was realized when species-specific equations were applied. Replacement of diameter-distribution data by quadratic mean diameter for each species reduced the absolute value of uncertainty, but created a bias when compared to the base case. Our attempt to incorporate regression standard error produced extremely large uncertainties for some equations and therefore was not pursued further. Results indicate that the most substantial reductions in uncertainty could be obtained by accurately assigning individual trees to suitable equations. The magnitude of model error produced by our methods currently precludes determination of significant differences between live-tree C stores of most landowners in the Pacific Northwest, and renders impossible the precise determination of the amount of live-tree C in a given forest area. Nevertheless, this does not necessarily preclude meaningful comparisons of C flux. Results of this study indicate uncertainty from model error in live-tree C could be extremely large. However, by accurately assigning appropriate volume or biomass prediction equations to trees, uncertainty could be greatly reduced.