Most of the seasonal channels in Lanzhou area do not have corresponding hydrological data, and it is impossible to obtain accurate flood peak flow, which brings a lot of inconvenience to channel management and management. Therefore, it is of great significance to establish a design flood peak flow calculation method for small watersheds without hydrological data in the loess area for disaster prevention and mitigation. Taking the mountain torrent channel without hydrological data in Lanzhou area of Gansu Province as the research object, combined with the survey and evaluation data of the Institute of Geological Natural Disaster Prevention and Control of Gansu Academy of Sciences, the consistency of the characteristic parameters of the channel basin and the correlation between the characteristic parameters and the design flood calculation were tested and analyzed by Kendall synergy coefficient test and grey correlation analysis. Two characteristic parameters of channel area and main channel length were selected to establish an empirical formula for calculating the design flood peak flow of mountain torrent channel in Lanzhou City. Combined with the comparative analysis of Lanzhou flood empirical formula method and 'Tie Yi Yuan' method, the mean errors of the three calculation methods are as follows: the average relative error of the new empirical formula method-Lanzhou flood empirical formula is about 14.4%, the average relative error of the new empirical formula method-'Tie Yi Yuan' method is about 17.8%, and the average relative error of the Lanzhou flood empirical formula method-'Tie Yi Yuan' method is about 20.3%. The calculation results of the new empirical formula are between the Lanzhou flood empirical formula and the 'Tie Yi Yuan' method. It has certain accuracy and rationality, and can be more efficient and convenient to calculate the design flood peak flow. At the same time, the new empirical formula deduces that the design flood peak flow only involves two characteristic parameters and can be accurately quantified.