Climate model projections show greater uncertainty for runoff (precipitation minus evapotranspiration) than for precipitation. Why is that? We investigate regional runoff sensitivities – how much runoff changes with warming and precipitation – in climate models and find that models have a widespread in runoff sensitivity. A model’s runoff sensitivity can explain directly how much runoff change a model predicts for the future, so this is clearly an important parameter for future projections. We gather observational estimates of runoff sensitivity and ask ‘which models have a realistic runoff sensitivity?’ We then constrain the future projections by using model projections together with observed sensitivity, basically giving weight to models that are better performing than others. This results in a reduction of uncertainty surrounding future projections, which is a result that is useful for both the modeling and stakeholder communities: the modeling community is made aware of systematic biases in climate models that should be focused on in future model development and, in the interim, stakeholders are served future runoff projections with reduced uncertainty.