Abstract. We introduce a 3-step statistical bias correction method to solve global climate model (GCM) bias by determining regional variability through multi-model selection. This is a generalized method to assess imperfect GCMs that are unable to simulate distinct regional climate characteristics, either spatially or temporally. More remaining local bias is eliminated using an all-inclusive statistical bias correction addressing the major shortcomings of GCM precipitation. First, the multi-GCM choice is determined according to spatial correlation (Scorr) and root mean square error (RMSE) of regional and mesoscale climate variation in a comparison with global references. After multi-GCM selection, there are three major steps in the proposed bias correction, i.e., the generalized Pareto distribution (GPD) for extreme rainfall bias correction, ranking order statistics for wet and dry day frequency errors, and a two-parameter gamma distribution for monthly normal rainfall bias correction. Best-fit GPD parameters are resolved by the RMSE, a Hill plot, and mean excess function. The capability of the method is examined by application to four catchments in diverse climate regions. These are the Kalu Ganga (Sri Lanka), Pampanga, Angat and Kaliwa (Philippines), Yoshino (Japan), and Medjerda (Tunisia). The assumption of GCM stationary error in the future is verified by calibration in 1981–2000 and validation over 1961–1980 at two observation sites. The results show a favorable outcome. Overall performance of catchments was good for bias-corrected extreme events and inter-seasonal climatology, compared with observations. However, the suggested method has no intermediate grid scale between the GCM grid and observation points, and requires a well-distributed observed rain gauge network to reproduce a reasonable rainfall distribution over a target basin. The results of holistic bias correction provide reliable, quantitative and qualitative information for basin or national scale integrated water resource management.