The large spatial and temporal variability of atmospheric aerosol load makes it a challenge to quantify aerosol effect on climate. This study is one of the first attempts to apply data assimilation for the analysis of global aerosol distribution. Aerosol optical thickness (AOT) observed from the Polarization and Directionality of the Earth Reflectances (POLDER) spaceborne instrument are assimilated into a three-dimensional chemistry model. POLDER capabilities to distinguish between fine and coarse AOT are used to constrain them separately in the model. Observation and model errors are a key component of such a system and are carefully estimated on a regional basis using some of the high-quality surface observations from the Aerosol Robotic Network (AERONET). Other AERONET data provide an independent evaluation of the a posteriori fields. Results for the fine mode show improvements, in terms of reduction of root-mean-square errors, in most regions with the largest improvements found in the Mediterranean Sea and Eurasia. We emphasize the results for the Arctic, where there is growing evidence of a strong aerosol impact on climate, but a lack of regional and continuous aerosol monitoring. The a posteriori fields noticeably well reproduce the winter-spring "Arctic Haze" peak measured in Longyearbyen (15¿E, 78¿N) and typical seasonal variations in the Arctic region, where AOT increase by up to a factor of three between a posteriori and a priori. Enhanced AOT are found over a longer period in spring 2003 than in 1997, suggesting that the large Russian fires in 2003 have influenced the Arctic aerosol load.