Classical approaches for flood risk assessment relate flood damage for a certain class of objects to the inundation depth, while other characteristics of the flooding situation and the flooded object are widely ignored. Observations on several discrete and continuous variables collected after the 2002 and 2005/2006 floods in the Elbe and Danube catch-ments in Germany offer a unique data mining opportunity in terms of learning a Bayesian Network. We take an entirely data-driven stance opting not to discretize continuous vari-ables in advance; rather, we cast the problem in Bayesian framework, and consider the maximum aposteriori of the joint distribution of the triple, network structure, parameters and discretization, as the outcome of the analysis. Moreover, motivated by the work of Merz et al. (2010), who point out the need of an improved flood damage assessment, we re-define the discretization of the target variable, flood loss, once the network has been learned. Its domain is split into a large number of intervals and the associated parameters are estimated using a Gaussian kernel density estimator. Although the prediction of the relative flood loss is comparable to state-of-the-art methods, our approach benefits from capturing the joint distribution of all factors influencing flood loss.