We analyzed the magnitude, the trends and the uncertainties of fossil-fuel CO2 emissions in the European Union 25 member states (hereafter EU-25), based on emission inventories from energy-use statistics. The stability of emissions during the past decade at EU-25 scale masks decreasing trends in some regions, offset by increasing trends elsewhere. In the recent 4 years, the new Eastern EU-25 member states have experienced an increase in emissions, reversing after a decade-long decreasing trend. Mediterranean and Nordic countries have also experienced a strong acceleration in emissions. In Germany, France and United Kingdom, the stability of emissions is due to the decrease in the industry sector, offset by an increase in the transportation sector. When four different inventories models are compared, we show that the between-models uncertainty is as large as 19% of the mean for EU-25, and even bigger for individual countries. Accurate accounting for fossil CO2 emissions depends on a clear understanding of system boundaries, i.e. emitting activities included in the accounting. We found that the largest source of errors between inventories is the use of distinct systems boundaries (e.g. counting or not bunker fuels, cement manufacturing, nonenergy products). Once these inconsistencies are corrected, the between-models uncertainty can be reduced down to 7% at EU-25 scale. The uncertainty of emissions at smaller spatial scales than the country scale was analyzed by comparing two emission maps based upon distinct economic and demographic activities. A number of spatial and temporal biases have been found among the two maps, indicating a significant increase in uncertainties when increasing the resolution at scales finer than ≈200 km. At 100 km resolution, for example, the uncertainty of regional emissions is estimated to be 60 g C m−2 yr−1, up to 50% of the mean. The uncertainty on regional fossil-fuel CO2 fluxes to the atmosphere could be reduced by making accurate 14C measurements in atmospheric CO2, and by combining them with transport models.