Twenty-two simulations from nine global coupled climate models were assessed for their ability to reproduce observed trends in a set of indices representing temperature and precipitation extremes over Australia. Observed trends for 1957-1999 were compared with individual and multi-modelled trends calculated over the same period. When averaged across Australia the magnitude of trends and interannual variability of temperature extremes were well simulated by most models particularly for the warm nights index. The majority of models also reproduced the correct sign of trend for precipitation extremes although there was much more variation between the individual model runs. A bootstrapping technique was used to show that most models produce plausible trends when averaged over Australia although only heavy precipitation days simulated from the multi-model ensemble showed significant skill at reproducing the observed spatial pattern of trends. Two of the models with output from different forcings showed that the observed trends for some of the temperature indices were inconsistent with the trends produced from the natural-only forcings runs but the forcing made little difference to the models' ability to reproduce the spatial pattern of trends over Australia. Future projected changes in extremes using three emissions scenarios were also analysed. Changes in temperature extremes were generally found to scale with the strength of emissions. Australia shows a shift towards significant warming of temperature extremes with much longer dry spells interspersed with periods of increased extreme precipitation irrespective of the scenario used.