Using results from the 4th Assessment Report of the Intergovernmental Panel on Climate Change, changes in the 20-year return levels for Australian daily maximum and daily minimum temperature were derived using extreme value theory. The climate models were evaluated using each of (a) mean performance, (b) skill in matching the observed probability density function and (c) skill in capturing the tails of the probability density function. Each weak-skilled model ensemble projected larger increases in both the maximum and minimum temperature return levels compared to each strong-skilled ensemble. The weak and strong skilled ensembles in maximum temperature were statistically significantly different. Over Australia, weakskilled models therefore simulated statistically significantly larger amounts of warming compared to stronger models in maximum temperature and larger amounts of warming in minimum temperature irrespective of how skill was assessed.