We use tropospheric NO2 columns from the Global Ozone Monitoring Experiment (GOME) satellite instrument to derive top-down constraints on emissions of nitrogen oxides (NOx ≡ NO + NO2), and combine these with a priori information from a bottom-up emission inventory (with error weighting) to achieve an optimized a posteriori estimate of the global distribution of surface NOx emissions. Our GOME NO2 retrieval improves on previous work by accounting for scattering and absorption of radiation by aerosols; the effect on the air mass factor (AMF) ranges from +10 to -40% depending on the region. Our AMF also includes local information on relative vertical profiles (shape factors) of NO2 from a global 3-D chemical transport model (GEOS-CHEM); assumption of a globally uniform shape factor, as in most previous retrievals, would introduce regional biases of up to 40% over industrial regions and a factor of 2 over remote regions. We derive a top-down NOx emission inventory from the GOME data by using the local GEOS-CHEM relationship between NO2 columns and NOx emissions. The resulting NOx emissions for industrial regions are aseasonal, despite large seasonal variation in NO2 columns, providing confidence in the method. Top-down errors in monthly NOx emissions are comparable with bottom-up errors over source regions. Annual global a posteriori errors are half of a priori errors. Our global a posteriori estimate for annual land surface NOx emissions (37.7 Tg N yr-1) agrees closely with the GEIA-based a priori (36.4) and with the EDGAR 3.0 bottom-up inventory (36.6), but there are significant regional differences. A posteriori NOx emissions are higher by 50--100% in the Po Valley, Tehran, and Riyadh urban areas, and by 25--35% in Japan and South Africa. Biomass burning emissions from India, central Africa, and Brazil are lower by up to 50%; soil NOx emissions are appreciably higher in the western United States, the Sahel, and southern Europe.