Abstract There was a dramatic decrease in rainfall in the southwest of Australia (SWA) in the mid-1960s. A statistical method, based on the idea of analogous synoptic situations, is used to help clarify the cause of the drying. The method is designed to circumvent error in the rainfall simulated directly by a climate model, and to exploit the ability of the model to simulate large-scale fields reasonably well. The method uses relationships between patterns of various atmospheric fields with station records of rainfall to improve the simulation of the local rainfall spatial variability. The original technique was developed in a previous study. It is modified here for application to two four-member ensembles of simulations of the climate from 1870 to 1999 performed with the Parallel Climate Model (PCM). The first ensemble, called “natural,” is forced with natural variations in both volcanic activity and solar forcing. The second ensemble, called “full forcing,” also includes three types of human-induced forcing resulting from changes in greenhouse gases, ozone, and aerosols. The full-forcing runs provide a better match to observational changes in sea surface temperature in the vicinity of SWA. The observed rainfall decline is not well captured by rainfall changes simulated directly by the model in either ensemble. There is a hint that the fully forced ensemble is more realistic, but it is nothing more than a hint. The downscaling approach, on the other hand, provides a much more accurate reproduction of the day-to-day variability of rainfall in SWA than the rainfall simulated directly by the model. The downscaled ensemble mean rainfall in full forcing declines over the region with a spatial pattern that is similar to the observed decline. This contrasts with an increase of rainfall in the downscaled rainfall in the natural ensemble. These results give the clearest indication yet that anthropogenic forcing played a role in the drying of SWA. Note, however, that ambiguities remain. For example, although the observed decline fits within the range of downscaled model simulation, the ensemble mean rainfall decline is only about half of the observed estimate, the timing differs from the observations, drying did not occur in the downscaling of one of the four full-forced ensemble members, and not all potential forcing mechanisms are included in full forcing (e.g., land surface changes). Furthermore, while the observed rainfall decline was a sharp reduction in the 1960s, followed by a near-constant rainfall regime, the full-forcing ensemble suggests a more gradual rainfall decline over 40 yr from 1960.