Rainfall forecast information from dynamical climate models can be harnessed for hydrological modelling via statistical post-processing that produces ensembles with reduced bias, less error and improved reliability in ensemble spread. In this study, a new Bayesian forecast calibration method called Ensemble Link Functions (ELFs) is proposed. It is an ensemble model output statistics type method that is designed to be generically applicable to post-process a) forecasts with both exchangeable and non-exchangeable ensemble members (e.g. when members are generated at multiple initiation times), b) hindcasts and forecasts being different in ensemble configuration, and c) a varying degree of correspondence between ensemble spread and real forecast uncertainty. We apply the new ensemble forecast calibration method to post-process rainfall forecasts from the Australian Bureau of Meteorology’s ACCESS-S model, which poses these challenges for post-processing. The complex distribution of rainfall is handled by embedding data transformation and left-censoring of rainfall values into the methodology. The new ensemble forecast calibration method is shown to be effective for producing reliable, bias-corrected forecasts of seasonal rainfall.