Conventional variational analysis schemes typically assume that the prior estimate has no systematic errors. This "perfect model" assumption is severely challenged in assimilating observation types such as precipitation for which the forward model is based on parameterized physics that are less than perfect. We present a variational data assimilation procedure that makes effective use of 6-h averaged rainrates derived from TRMM and SSM/I microwave sensors to improve forecasts and analyses produced by the Goddard Earth Observing System (GEOS) Data Assimilation System (DAS). The procedure minimizes the least-square differences between the observed rainrates and those generated by a colume model with prescribed forcing. The scheme generates moisture and temperature tendency corrections to compensate for apparent errors arising from model deficiencies and initial conditions. We show the impact of rainfall assimilation using this procedure in the current versions of the GEOS DAS.