 This paper presents a technique for assessing the predictive uncertainty of rainfall runoff and hydraulic forecasts. It uses quantile regression to condition forecast uncertainty on the forecasted value itself based on retrospective analysis of hind-casted water level forecasts and forecast errors. This allows us to estimate the probability distribution of forecast errors given the forecasted water levels. We tested this technique with data from various catchments in England and Wales and found it to be effective and robust. This article was authored by A. H. Wirts, H. C. Wincemias and J. S. Vercade.