Knowledge of NOx (NO2 + NO) emissions is useful to understand processes affecting air quality and climate change. Emission inventories of surface NOx have high uncertainties. Satellite remote sensing has enabled measurements of trace gases in the atmosphere over a large regional and temporal scale. Inverse modeling of NO2 observations from satellites can be used to improve existing emissions inventories. This study seeks to understand the difference in two methods of inverse modeling: the mass balance approach and the adjoint approach using the GEOS-Chem chemical transport model and its adjoint. Using both synthetic satellite observations and those derived from the SCIAMACHY satellite instrument, this paper found that the performance of these two inversions was affected by pixel smearing and observational error. Smearing reduced the accuracy of the mass balance approach, while high observational error reduced the accuracy of the adjoint approach. However, both approaches improved the a priori emissions estimate.