Under the United Nations Framework Convention for Climate Change all Parties have to report on carbon emissions and removals from the forestry sector. Each Party can use its own approach and country specific data for this. Independently, large-scale models exist (e.g. EFISCEN and G4M as used in this study) that assess emissions and removals from this sector by applying a unified approach to each country, still often based on country specific data. Differences exist between the national reported values and the calculations from the large scale models. This study compares these models with national reporting efforts for 24 EU countries for the period 2000-2008, and identifies the most likely causes for differences. There are no directly identifiable single input parameters that could be targeted to fully close the gap between country and model estimates. We found that the method applied by the country (i.e. stock-difference or gain-loss) contributes significantly to differences for EFISCEN and was the best explaining variable for G4M, although for the latter it was not significant. Other variables (biomass expansion factors, harvest volumes and the way harvest losses are treated) were not found to provide a conclusive explanation for the differences between the model estimations and the country submissions in an over-all analysis. However, at the level of individual countries several different causes for differences were identified. This suggests that to really close the gap between country submissions and large scale models, close collaboration between modellers and country experts is needed, calling for openness and willingness to share relevant data and to compare GHG inventories with independent estimates. This would enable to improve the confidence both in historical GHG inventories and in the models which are needed to project the future forest sink for several policy issues.