The adequacy of current passive-microwave-(PM)- and infrared-(IR)-based satellite rainfall retrieval and sampling for flood prediction of a medium-sized watershed is investigated. On the basis of Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar rainfall measurements, rain retrieval error parameters for PM and IR sensors are derived. PM rain retrievals are inferred from the overland component of the TRMM Microwave Imager (TMI) rain estimation algorithm, while IR retrievals are obtained from hourly PM-calibrated IR rain fields, which are part of a variable rainfall product (VAR) array produced at NASA/GSFC. A probabilistic error model is developed for satellite-based precipitation measurements on the basis of retrieval error parameters in this simulation study. The PM rain detection ability was found to be significantly more sensitive than that of IR while the successful no-rain detection probabilities were found to be 93% and 88%, respectively. The IR retrieval was found to give false alarm rain rates about twice as large as that of PM. The PM sensor constellation comprised two Special Sensor Microwave Imagers (SSM/I) (F14 and F15), the TMI, and the Advanced Microwave Sensing Radiometer (AMSR-E). It was found that current PM sampling is associated with flood prediction uncertainty approximately 50--100% higher than that of a canonical 3-hourly sampling planned for the Global Precipitation Measurement (GPM) mission. The comparatively greater limitation in capturing the correct space-time rain structure by IR retrievals had the effect of increasing the error in predicting the time of peak runoff when merging was performed with PM retrievals. It was found that a reduced standard error (0.90) can make IR retrievals useful in reducing uncertainty in the prediction of peak runoff. To reduce the error in time to peak, further improvement, such as reduction in the IR retrieval's false alarm rates coupled with an even higher POD, may be necessary. In terms of overall runoff volume, combined moderate improvements in POD and error variance of current IR retrieval algorithms are sufficient for the reduction of prediction uncertainty.