Statistical methods used to construct future climate projections for hydrological modeling are commonly validated against the observed climate. Here, we assess the relative performance of nine projection methods for daily precipitation directly in the future climate using intermodel cross-validation. Regional climate models used in the study were selected from the ENSEMBLES data archive. The results show that bias correction methods tend to outperform delta change methods in northern Europe. In particular, bias correction using quantile mapping often has the best cross-validation statistics. However, none of the methods performs universally well, which indicates that part of the uncertainty related to method selection is unavoidable. To test to what extent method differences need to be taken into account in projection uncertainty assessments, the relative importance of method and model uncertainties was quantified using analysis of variance. Although model differences generally explain larger fraction of the spread in the precipitation projections, contribution of method uncertainties is non-negligible, especially in the low and high intensity parts of the distribution. Thus, several methods should be used in parallel when assessing future climate changes. Uncertainties in climate model projections are further transferred to hydrological modeling. Their effect on simulated hydrological changes in Scandinavian region will be assessed using HYPE (Hydrological Prediction for the Environment) hydrological model. We will briefly introduce the model and the ongoing work carried out in the Academy of Finland funded RECAST project.