Abstract: Better knowledge of sediment concentrating processes for reservoir dam sites is crucial in reservoir water and sediment operations.In this study,a flood similarity-based method is developed for predicting reservoir sediment concentrations in heavily sediment-laden rivers.The method uses the main principle of the total flow model for sandy rivers.An approach based on the cluster analysis of historical flood events is employed in the model application.The events are first grouped and then used separately in calibrating the similarity-based method.Thus,the same group flood events will always share the same set of model parameter values in the model applications,and the idea is also applicable to the future flood events.Model validation results show that the sediment concentrating process of reservoir dam-sites can be better predicted with the grouped parameter values.The improvement is especially significant in predicting the peak value of sediment concentrations.The similarity-based method could be potentially useful in the decision making process of reservoir sediment flushing.