A flash flood early warning method based on dynamic critical precipitation is proposed in this paper. The method calculates soil moisture saturation using the Xin’anjiang model, then sets up forewarning discriminant functions for three time scales, 6 h, 12 h and 24 h, using Widrow Hoff algorithm of minimum mean square error criteria, and based on soil moisture saturation and maximum 6 h, 12 h and 24 h precipitation before each flood event. With this method, together with rainfall data of the 17 flood events occurring during the period from 2003 to 2009, wich are obtained at precipitation stations in Pihe Basin parameters for the model are calibrated and the method is applied in 10 historical flash flood events to verify the three time scale forewarning discriminent functions. The correct forecast rate is over 70%, demonstrating that the method is reliable and can be used to flash flood forecasting and early warning.