As part of a collaborative project between NCEP and investigators NOAA's Earth System Research Lab (ESRL), the University of Maryland (UMD) and the Colorado State University (CSU), several different formulations of the Ensemble Kalman filter (EnKF) have been tested within the NCEP Global Forecast System. All non-radiance observations for January and February 2004 were assimilated, using a reduced resolution (T62) version of the global forecast model operational at that time. The purpose of this project was to see whether ensemble data assimilation is a viable candidate for NCEP's next generation data assimilation system, and to see if there are any quantitative differences between the different EnKF formulations developed by the different research groups. In this presentation, I will focus on the implementations developed at ESRL at UMD. Comparisons of both algorithms with a benchmark run of the operational 3DVar system run at NCEP will be presented.