Bias‐correction methods applied to monthly temperature and precipitation data simulated by multiple General Circulation Models (GCMs) are evaluated in this study. Although various methods have been proposed recently, an intercomparison among them using multiple GCM simulations has seldom been reported. Moreover, no previous methods have addressed the issue how to adequately deal with the changes of the statistics of bias‐corrected variables from the historical to future simulations. In this study, a new method which conserves the changes of mean and standard deviation of the uncorrected model simulation data is proposed, and then five previous bias‐correction methods as well as the proposed new method are intercompared by applying them to monthly temperature and precipitation data simulated from 12 GCMs in the Coupled Model Intercomparison Project (CMIP3) archives. Parameters of each method are calibrated by using 1948–1972 observed data and validated in the 1974–1998 period. These methods are then applied to the GCM future simulations (2073–2097) and the bias‐corrected data are intercompared. For the historical simulations, negligible difference can be found between observed and bias‐corrected data. However, the differences in future simulations are large dependent on the characteristics of each method. The new method successfully conserves the changes in the mean, standard deviation and the coefficient of variation before and after bias‐correction. The differences of bias‐corrected data among methods are discussed according to their respective characteristics. Importantly, this study classifies available correction methods into two distinct categories, and articulates important features for each of them.