Avalanche disasters are associated with significant monetary losses. It is thus crucial that avalanche risk assessments are based on a consistent and proper assessment of the uncertainties involved in the modelling of the avalanche run-out zones and the estimations of the damage potential. We link a Bayesian network (BN) to a Geographic Information System (GIS) for avalanche risk assessment in order to facilitate the explicit modelling of all relevant parameters, their causal relations and the involved uncertainties in a spatially explicit manner. The suggested procedure is illustrated for a case study area (Davos, Switzerland) located in the Swiss Alps. We discuss the potential of such a model by comparing the risks estimated using the probabilistic framework to those obtained by a traditional risk assessment procedure. The presented model may serve as a basis for developing a consistent and unified risk assessment approach.