The present and doubled CO2 equilibrium climates simulated by slab ocean versions of the atmospheric general circulation models from the Commonwealth Scientific and Industrial Research Organisation (CSIRO, Mark 1 and Mark 2) and from the Bureau of Meteorology Research Centre (BMRC) are examined, with the aim of explaining the large variation in mean warming (4.8°C, 4.3°C, and 2.1°C). The present climates are compared firstly with observations. A graphical display of nondimensional measures of local and mean errors is used. For 15 quantities the models produce broadly similar skill, which indicates that such an evaluation is of limited use as a validation of these models for climate change prediction. Comparison of the two climates indicates that for temperature, snow/ice cover, and water column (but not necessarily other fields) the typical magnitudes of local changes are in rough proportion to the mean warming. For tropical precipitation, however, the BMRC model shows a similar sensitivity to CO2 doubling as do the CSIRO models. A standard diagnostic feedback analysis shows that the Mark 1 model has stronger albedo, water vapor, and cloud feedbacks than the BMRC model. A novel regional net feedback analysis is then applied to all three models. Feedbacks for the snow/ice region and clear-sky and cloud forcing components of the snow-free region indicate similar intermodel differences to those from the diagnostic approach. The feedbacks are examined in relation to the simulated climates and model parameterizations. As the application of the regional method requires only standard climatological fields, it is proposed as a convenient analysis tool in further model comparisons.