An experimental ensemble data assimilation component for the NCEP global forecast system (GFS) has been developed as part of a collaboration between NCEP, NOAA/ESRL and the University of Maryland. This experimental system produces more accurate forecasts of geopotential height and mean sea level pressure than the current operational 3D-Var system run at the same resolution when no satellite-derived brightness temperature information is assimilated. The biggest improvements are seen where observations are sparse (in the Southern Hemisphere and above the tropopause). In this study, we will focus on the impact of satellite brightness temperature observations on the ensemble data assimilation system. Technical issues related to the assimilation of satellite radiance information in the ensemble Kalman filter will be discussed, and the accuracy of the assimilated states will be assessed by comparing the skill of forecasts initialized by the experimental system and the current NCEP operational system run at the same resolution. Of particular interest is whether the advantage that the experimental ensemble-based system had when only ‘conventional' observations were assimilated carries over when all available observations are assimilated.