The emission of mineral dust aerosols in arid and semiarid regions is a complex process whose representation in atmospheric models remains crude, due to insufficient knowledge about the aerosol lifting process itself, the lack of global data on soil characteristics, and the impossibility for the models to resolve the fine‐scale variability in the wind field that drives some of the dust events. As a result, there are large uncertainties in the total emission flux of mineral dust, its natural variability at various timescales, and the possible contribution from anthropogenic land use changes. This work aims for estimating dust emissions and reduces their uncertainty over the Sahara desert and the Arabian Peninsula—the largest dust source region of the globe. We use a data assimilation approach to constrain dust emission fluxes at a monthly resolution for 18 subregions. The Moderate Resolution Imaging Spectroradiometer satellite‐derived aerosol optical depth is assimilated in a regional configuration of a general circulation model coupled to an aerosol model. We describe this data assimilation system and apply it for 1 year, resulting in a total mineral dust emissions flux estimate of 2900 Tg yr−1 over the Sahara desert and the Arabian Peninsula for the year 2006. The analysis field of aerosol optical depth shows an improved fit relative to independent Aerosol Robotic Network measurements as compared to the model prior field.