Relevance. The Amur basin is situated in a flood-inclined region. Over the course of the past century the disastrous pluvial flooding have occurred in the basin eight times. They covered huge territories, took peopleґs lives and caused considerable damage to residential and industrial buildings, engineering systems. One of the efficient methods to struggle the floods is to regulate the river run-off developing flood-control reservoirs. The number of existing reservoirs on the Amur river system is insignificant to control efficiently the river run-off. Their building is tied with further region development. The forecast of flood water inflow allows minimizing damage and identifying the most rational reservoir release rules for the existing and expected reservoirs. Aim of the research is to investigate the floods formation by the example of the Onon river and to develop the methods for short-term forecast of floods in the upper Amur basin. Research methods: water balance method, geographical and hydrological methods, statistical method, mathematical modeling of run-off formation. Results. The Burakovґs conceptual model is adapted to forecast daily rain floods water flows. This model is used by the Siberian department of the Federal Service for Hydrometeorology and Environmental Monitoring of Russia. The landscape and hydrological basis for this model is the basin division into areas and altitudinal zones. The altitudinal zone is an initial territorial unit of averaging the hydrological characteristics of mountain rivers. Within the altitudinal zone, territorial irregularity of snow cover distribution and capacitive water absorption are taken into account by probability distribution. The periods, when the diurnal water entry to the surface of the basin exceeds the diurnal evaporation and infiltration, compose successive flood-forming periods. For each day of a flood-forming period, the water yield is estimated using the Popovґs infiltration capacitive model. Besides, the gravitational water storage on the slopes and the inflow in the channel network are calculated. The method, describing water lag along the river channels, is based on applying the convolution integral (the genetic flood formula). As a result of the research, the author has developed the method of forecasting daily water levels in the Onon riverbed system. The method was applied by the Chita department of the Federal Service for Hydrometeorology and Environmental Monitoring of Russia and proved its efficiency.