The lag time between rainfall input and discharge output-that is, the response speed of a catchment-is a key parameter for flood prediction but is difficult to quantify. Here we propose a new method to quantify catchment-wide average peak propagation speed in hillslopes based on measurements conducted in a steep 4.5-km(2) catchment. The peak lag time of the catchment was partitioned into lag times in the hillslope and channel. Peak propagation speed was calculated from flow path length divided by lag time. We verified the results and evaluated the applicability and uncertainty of this method. This method can be applied during intense storms with identifiable peak relationships between rainfall and discharge. The effect of spatial variation in precipitation was insignificant in the study catchment. The increase in rainfall intensity and the rainfall accumulated prior to peak rainfall increased peak propagation speed. The method used to characterize flow path length from distributed values introduced some uncertainty. We calculated modal lengths using linear and logarithmic scales and various bin widths. These calculations resulted in small differences in modal channel length but led to more than tenfold differences in modal hillslope length. Thus, uncertainties for peak propagation speed across all hillslopes varied by one order of magnitude. Hillslope lag times were less than 2 min and peak propagation speed in the hillslopes ranged from 0.04 to 4.3 m/s for floods with recurrence intervals of one to several years. The proposed method and calculated data should improve prediction accuracy for extreme floods.