The Bayesian probability inversion and a Markov chain Monte Carlo (MCMC) technique were applied to a terrestrial ecosystem model to analyze uncertainties of estimated carbon (C) transfer coefficients and simulated C pool sizes. This study used six data sets of soil respiration, woody biomass, foliage biomass, litterfall, C content in the litter layers, and C content in mineral soil measured under both ambient CO2 (350 ppm) and elevated CO2 (550 ppm) plots from 1996 to 2000 at the Duke Forest Free-Air CO2 Experiment (FACE) site. A Metropolis-Hastings algorithm was employed to construct a posterior probability density function (PPDF) of C transfer coefficients on the basis of prior information of model parameters, model structure, and the six data sets. The constructed PPDFs indicated that the transfer coefficients from pools of nonwoody biomass, woody biomass, and structural litter were well constrained by the six data sets under both ambient and elevated CO2. The data sets also gave moderate information to the transfer coefficient from the slow soil C pool. However, the transfer coefficients from pools of metabolic litter, microbe, and passive soil C were poorly constrained. The poorly constrained parameters were attributable to either the lack of experimental data or the mismatch of timescales between the available data and the parameters to be estimated. Cumulative distribution functions were constructed for simulated C pool sizes on the basis of the six data sets, showing that on average the ecosystem would store 16,616 g C m-2 at elevated CO2 by the year 2010, significantly higher than 13,426 g C m-2 at ambient CO2 with 95% confidence. This study shows that the combination of a Bayesian approach and MCMC inversion technique is an effective method to synthesize information from various sources for assessment of ecosystem responses to elevated CO2.