The flash flood early warning method based on dynamic critical precipitation is proposed, which takes into account the percentage saturation of soil moisture content in double-excess model. A series of historical precipitation data of various gauge stations in the upper catchment of the study area at the early warning cross-section are set as the input parameters, thereby the runoff generation and concentration in the catchment are obtained in the double-excess model, and the percent saturation of soil moisture content is calculated. Based on the warning discharge in combination with the percentage saturation of soil moisture content, the discriminant relations of the critical precipitation for the time intervals, including 0.5 h, 1 h, 1.5 h, 2 h, 2.5 h, and 3 h, are computed respectively using the inversion method. Using the precipitation data from ground rain gauge stations for year x and flood hydrograph data of x typical flood events for the Dayuhe River catchment, the SCE-UA algorithm is adopted to calibrate the parameters of the double-excess model, and the discriminant functions of dynamic critical precipitation for flash flood early warning with 6 time scales are validated using x representative historical flood hydrographs. The qualification ratio for flash flood early warning exceeds x, which demonstrates the feasibility and applicability of the proposed method.