Formation processes of the runoff of rainfall floods are considered and the short-term forecast technique is developed by using the Onon river as an example, because the hydrological regime of the rivers within the Upper-Amur basin are characterized by significant water-level fluctuations caused by rainfall floods during the summer-autumn period. The investigation is based on the water-balance method. Plots of changes of the water flow rate over time (hydrographs) are constructed, and genetically related values of the flood runoff and flood-forming precipitation are calculated for the period of long-term observations. Statistical characteristics are used to substantiate the forecast arguments (predictors): the pair and multiple correlation coefficients and the criteria of t-statistics which are numerically equal to the ratio of the regression coefficient to the standard error of its determination. It is established that in addition to atmospheric precipitation that makes the main contribution to the formation of the flood runoff, an important role is also played by the degree of soil moistening prior to flood-forming rainfall. It is suggested that the water level in the river prior to the onset of a flood should be used as the indicator of pre-flood moistening. Two approaches are tested in dealing with the problem of forecasting the volume and the maximum water level of rainfall floods. E. G. Popov’s infiltration-capacity model was used to determine the mathematical structure of the dependence of the flood runoff on the main factors for its formation having regard to the main components of melt water losses (capacity retardation, infiltration, evaporation). The resulting dependences are recommended for making forecasts of the volume and the maximum water level of rainfall floods on the rivers within the Onon basin.Keywords: flood runoff, water balance, runoff formation models, hydrological forecasts.