We analyze changes in flood hazard in Europe by examining extreme discharge levels as simulated by the hydrological model LISFLOOD when driven by a multimodel ensemble of climate simulations. The ensemble consists of simulations from two regional climate models (RCMs), both run with boundary conditions from two global models, and for two scenarios of greenhouse gas emissions. In northeastern Europe, a general decrease in extreme river discharge is observed in the scenario period, suggesting a reduction in the hazard of extreme snowmelt floods. Elsewhere, we find a consistent tendency toward a higher flood hazard in the majority of the model experiments in several major European rivers. These changes can partly be attributed to large, decadal-scale variability in the simulated climate and can be expected to occur naturally when comparing two 30-year time periods, even without a change in greenhouse gas forcing. We furthermore find evidence for a considerable influence of especially the global model that is used to drive the RCMs. At the scale of individual river basins, using a different combination of climate models or assuming a different emissions scenario sometimes results in a very different or even opposite climate change signal in flood hazard. We therefore believe that a multimodel approach as adopted in the present paper provides the best way to address the various uncertainties in impact studies of hydrometeorological extremes. Probabilistic scenarios that consist of multiple realizations of the current and future climate state are indispensable to better identify the climate signal amidst large variability.