Abstract: A probabilistic flood forecasting method considering the heterogeneity in error distributions of deterministic predictions is presented. In this method, error distributions of different flood magnitudes are constructed using their observed values and model forecasting values (MFVs) and their probabilistic distribution functions (PDFs) conditional on the MFVs are derived; then, by using these PDFs a probabilistic flood forecasting is achieved. In its application to the Wangjiaba station in the Huai River basin, an empirical model was adopted in deterministic flood predictions of the ten flood events occurring in the period of 1996-2007, and variations in the relative errors of the deterministic predictions were examined. This analysis revealed that the means of PDFs of these relative errors differ significantly between different flood magnitudes. To consider such heterogeneity of error distributions, we can construct a functional relationship of the mean value of the relative errors versus their corresponding MFVs, and derive error distributions conditional on the MFVs. This new method was verified by using the data of two floods occurring in 2008. Results show that the probabilistic method is simple and practical and it can give more accurate forecasts than the empirical model when the means of PDFs of discharges are used as deterministic predictions.