Precipitation frequency (PF) analysis from global with regional climate model (GCM+RCM) simulations provides useful information about extreme precipitation events. Despite being useful tools for developing regional-to-local climate data, GCM+RCMs suffer from uncertainties due to imperfect physics, systematic errors from the driving GCMs, uncertainty in future scenarios, etc. Most multi-model historical and future analyses of PF estimates are based upon some kind of averaging (such as median) of PF estimates computed for each model individually. Such multimodel analyses have limitations: they do not assess the historical performances of models; they have large estimation uncertainty due to short integration length; and they tend to reduce the magnitude of extreme events due to averaging. We demonstrate a methodology for multimodel analysis of PF estimates that involves (i) historical assessment of GCM+RCMs, (ii) bias correction of the historical and future simulations of reasonably performing GCM+RCMs, and (iii) pooling of the bias corrected models. We apply this methodology to 13 CORDEX model simulations to quantify future changes in PF estimates using nonstationary generalized extreme value (GEV) analysis of annual maximum precipitation in the Susquehanna and Southern Florida watersheds. We show that our methodology is able to reduce estimation uncertainty and variability in PF estimates across models, and that the future signals are better resolved from their historical counterparts, enabling a clearer picture of significant changes in future PF estimates. We plan to repeat the analysis in newly-released CMIP6 models to compare performances of CORDEX models with high resolution state-of-the-art GCMs.