The optical and microphysical features of the cloud top are key information for quantitative precipitation retrieval based on satellite spectra. With the spectral and radar observation data of the Tropical Rainfall Measuring Mission satellite collected during the Meiyu periods of the Yangtze-Huaihe Valley from 1998 to 2007, this paper built a random forest algorithm model that used the cloud-top spectral information of precipitating clouds to retrieve precipitation intensity. It also investigated the relationship between cloud-top microphysical characteristics and the intensity variation of Meiyu precipitation. The results show that in the precipitation retrieval testing set of the random forest model, the correlation coefficient R between predicted precipitation intensity and observed precipitation intensity is 0.67, and the root-mean-square error is 4.06 mm/h. This means the random forest model has high precipitation prediction accuracy. In the model, the cloud water path (CWP) ranks high in the importance hierarchy of all input variables. Further analysis shows that when the CWP is leas than 1.0 kg·m ^-3, precipitation at light rain level is dominant during the Meiyu period of the Yangtze-Huaihe Valley, while when the CWP is more than 1.5 kg·m ^-3, the probability of precipitation at heavy rain and torrential rain levels increases significantly. On the whole, the probability of precipitation at all levels increases monotonically with the increase of cloud effective radius (CER), which is mainly over 10 μm. It also increases with the increase of cloud optical thickness (COT). When COT is more than 120, it increases significantly, especially the probability of heavy precipitation.