A new methodology for rainfall retrievals from indirect measurements ii pro
posed and illustrated using IR brightness temperature and radar rainfall ob
servations collected during TOGA-COARE. Since (1) rain rate has a mixed dis
tribution with a delta-function for a zero rain and lognormal distribution
for nonzero and (2) the least squares method which is used to calculate reg
ression coefficients gives a priori consistent estimates only for normally
distributed data. it is proposed to convert the rain rate to a normally dis
tributed set and only after that to develop a retrieval method and estimate
the skill of this method. Consideration of the physics of clouds and cloud
ensembles, the goal to minimize errors in the radar data, and the desire t
o remove the influence of cirrus clouds lead us to use: a) minimum of IR br
ightness temperatures over a 1 degrees x 1 degrees area and a 3 hour interv
al as a predictor, and b) radar rainfall, averaged over 3 hours over a 1 de
grees x 1 degrees area, with the radar in its center, as the truth. Results
using the TOGA-COARE data show that the correlation of the rain rate trans
formed to normal distribution is significantly higher with minimum temperat
ure than with the fraction of area covered by high clouds. The sizes of hea
vy rainfall areas obtained using the new methodology are reasonable. The re
gression coefficients should change with latitude, season and location. Tak
en together, the results indicate that it is possible, in principle, to ret
rieve rainfall from IR satellite observations and obtain reliable rainfall
data. To realize this goal it is necessary to process radar and IR data usi
ng the new methodology for different latitudes, seasons, over land and ocea
n.