Peak flow rate from watersheds is an important criterion used to develop soil conservation plans and to design engineering projects. A peak flow rate equation used in the CREAMS model, with four parameters, can be employed to predict peak flow rate. The purpose of this study was to test and improve this equation of peak flow rate in CREAMS for use on the Loess Plateau of China. Data from 331 storms in 20 small watersheds were used to verify the the peak flow rate equation in CREAMS. The calculated flow rates using the CREAMS equation greatly underestimated the measured peak flows. The model efficiency was only 0·15. Nonlinear regression analysis was then performed to develop a new equation: $$Q_{p} = 6{\cdot}69A^{0{\cdot}59}R^{1{\cdot}15 A^{0{\cdot}06}}P^{-0{\cdot}72}$$ which gave a model efficiency of 0·94. A second set of data, including 68 storms from four completely different watersheds, was used to test the new equation, with a resultant model efficiency of 0·90. The result has significant implications for improving the design of soil and water supporting practices, for assessing the soil and water resources, and for implementing conservation programmes. Copyright © 2008 John Wiley & Sons, Ltd.