This study focuses on the retrieval of liquid precipitation from the synergetic use of both TRMM Microwave Imager (TMI) and Precipitation Radar (PR). Rain rates are retrieved from TMI measurements with two different statistical approaches developed on the same database. This database is made of measured data obtained through co-localisation of the TMI brightness temperatures and the rain rates from PR. The main advantage of such a database is its representativeness which is theoretically perfect. An emerging statistical tool for remote sensing is the Self Organizing Map (SOM). Algorithms proposed by Kohonen have been used to obtain this map. SOM maps allow the projection of measured brightness temperatures into a two dimensional space. The obtained map is topologically ordered, that is to say: the spatial location of each neuron corresponds to a particular domain or feature of input data. SOM algorithm is thus applied to TMI measurements and the neurons have been labeled with corresponding PR rain rate. The labeled map is used for rain rate retrieval. A bayesian classifier is also used for comparison. It provides with a retrieved rain rate by calculating a weighted average of the rain rates stored in the database. The weighting is proportional to the distance in the brightness temperature space between the stored and the measured vector. A series of parameters are tuned because they are extremely difficult to estimate. After a comparison of these methods in terms of performances (bias and error variance), time computing and topology of the retrieved field the obtained rain rate are thus compared from two different points of view: Global maps with a precipitation product at 0.5 degree x 0.5 degree resolution averaged on a monthly basis. This will allow us to verify the absence of mean bias Case study corresponding to tropical situation (hurricanes and cyclones) will also be presented to test the quality of retrieved rain structures