A method is described for evaluating the ‘partial derivatives’ of globally averaged top-of-atmosphere (TOA) radiation changes with respect to basic climate model physical parameters. This method is used to analyse feedbacks in the Australian Bureau of Meteorology Research Centre general circulation model. The parameters considered are surface temperature, water vapour, lapse rate and cloud cover. The climate forcing which produces the changes is a globally uniform sea surface temperature (SST) perturbation. The first and second order differentials of model parameters with respect to the forcing (i.e. SST changes) are estimated from quadratic least square fitting. Except for total cloud cover, variables are found to be strong functions of global SST. Strongly non-linear variations of lapse rate and high cloud amount and height appear to relate to the non-linear response in penetrative convection. Globally averaged TOA radiation differentials with respect to model parameters are also evaluated. With the exception of total cloud contributions, a high correlation is generally found to exist, on the global mean level, between TOA radiation and the respective parameter perturbations. The largest non-linear terms contributing to radiative changes are those due to lapse rate and high cloud. The contributions of linear and non-linear terms to the overall radiative response from a 4 K SST perturbation are assessed. Significant non-linear responses are found to be associated with lapse rate, water vapour and cloud changes. Although the exact magnitude of these responses is likely to be a function of the particular model as well as the imposed SST perturbation pattern, the present experiments flag these as processes which cannot properly be understood from linear theory in the evaluation of climate change sensitivity.