The availability of long-term land datasets is fundamental to ensure consistent and accurate national greenhouse gas inventories for land use and land use changes. Estimation of Romania’s CO2 removals and emissions over 1970-2010 was assessed comparatively by using National Statistics (as reference dataset), Corine Land Cover data, as well as four additional datasets produced from the reference dataset and assuming some forest data changes according to other available partial datasets (NFI,LUCAS or expert judgment). A spreadsheet, implementing a model of UNFCCC national greenhouse gas inventory, allowed estimation of both net CO2 removals and emissions and gross CO2 fluxes from all carbon pools, for all land subcategories over 1970-2010. The model was run for each dataset. The reference dataset resulted in an average annual gross flux of CO2, twice as large as the corresponding annual net removals of 13 TgCO2, while annual estimates were more than double in post-1990 compared to pre - 1990. Uncertainty of net annual CO2 removals by reference dataset was estimated around 31 %, slightly lower than for gross estimate, and just around 55% when based on CLC dataset. Overall, the contribution of 12 broad land sub-categories to CO2 inventory was similar in the national net and gross estimates, for both pre- and post-1990 periods. Land under conversion represented 9% of the country’s area in pre- and only 2% in post-1990, corresponding to an annual average of 28% of gross and 6% of net annual CO2 estimates. Among the choices of datasets tested, the reference dataset provided the most conservative estimates of the CO2 inventory. Other datasets generally overestimated annual LULUCF sink and its main contributor forestland, as well as the emissions or removals from land conversion. Compared to pre-1990, when annual sink was rather low, land abandonment and moderating management in the post-1990 period has led to an increase of C stock in all pools, showing the relevance of political changes on land’s CO2 emissions/removals. Inconsistency within available land datasets impairs more accurate estimation of national GHG inventory. The development of an improved land use assessment system around National Forest Inventory is therefore suggested as a solution to implement consistent land definitions and to accurately estimate their areas in time.