Retrieval uncertainty estimates for vertical tropospheric NO2 columns based on theoretical error source discussions combined with actual Global Ozone Monitoring Experiment (GOME) observations are presented. Contributions to the total retrieval uncertainty are divided into three categories: (1) errors caused by measurement noise and spectral fitting, affecting the slant column density, (2) errors related to the separation of stratospheric and tropospheric NO2 affecting the estimate of the stratospheric slant column, and (3) errors due to uncertainty in model parameters such as clouds, surface albedo, and a priori profile shape, affecting the tropospheric air mass factor. Furthermore, it is shown that a correction for the effective temperature of the trace gas is essential and that a correction for the presence of aerosols needs to be accompanied by aerosol corrections to the cloud retrieval. A discussion of the error components and total retrieval uncertainty is given for March 1997. Tropospheric NO2 columns can be retrieved with a precision of 35--60% over regions with a large contribution of the troposphere to the total column. This error estimate demonstrates the need for highly accurate albedo maps, cloud retrieval schemes, and realistic a priori NO2 profile shapes.