In this study, TRMM-observed precipitation in the tropics is decomposed according to the horizontal area of radar precipitation features, with special emphasis on large systems (rain area > 104 km2) that contribute roughly half of tropical rainfall. Statistical associations of rain-weighted radar precipitation feature (RPF) size distributions with atmospheric variables on the 1.5° grid of ERA-Interim data are explored. In one-predictor distributions, the association with total precipitable water vapor (TPWV) is the strongest, while relative humidity at low and midlevels and low-level wind shear are also positively related to large-RPF rain fraction. Standard CAPE and CIN variables computed from grid-mean thermodynamic profiles are only weakly related to the size of rain systems. Joint distributions over two variables are also reported. The relative importance of predictors varies over different regions. The eastern Pacific is distinctive for having large rain systems in environments with a moist boundary layer but a dry midtroposphere, with strong shallow wind shear and small CAPE. In contrast, the large-storm environment over the western Pacific is found to be moister in the whole troposphere, with relatively weaker wind shear and larger CAPE. Over tropical land, the Sahel and central Africa stand out as having a great fractional rainfall contributed by large RPFs. Their associated environment is characterized by lower TPWV but stronger shallow wind shear and larger CIN and CAPE, in comparison to the equatorial Amazon basin and the Maritime Continent. Based on these associations, statistical reconstructions of the geographical distribution of large-RPF rain fraction from grid-mean atmospheric predictors are attempted.