This study considers an "offline" data assimilation method using the Ensemble Kalman Filter to build a precipitation reanalysis over France. The method is here applied for reconstructing the 2009-2012 period, using past observation density of 1871, 1900 and 1950. The methodology allows taking two main features of precipitation into account: (1) an anisotropic localization matrix based on the climatological background information, (2) a Gaussian transformation applied to daily precipitation. Results show a reduced error and a reduced uncertainty compared to background reconstructions, even with few observations, thus demonstrating the added value of data assimilation.