The response of a general circulation model (GCM) to global perturbations in sea surface temperatures (SSTs) is examined. The feedback strengths in the model are diagnosed by the response of top of atmosphere (TOA) radiative fluxes determined after substitution of fields from the perturbed climate into the control. Total feedback is divided into terms due to water vapour, lapse rate, surface temperature, and clouds (in turn analysed in terms of cloud amount, height and types). The standard experiment prescribes a globally uniform SST perturbation with fixed soil moisture. Four additional experiments vary the number of model vertical levels, the pattern of SST changes, the convection scheme, and the soil moisture. The SST pattern change chosen follows that of an equilibrium 2¿CO2 experiment, which shows polar amplification of the surface warming. Variations in the clear sky sensitivity of the model are shown to depend primarily on changes in the long wave response due to competing (positive) water vapor and (generally negative) lapse rate feedbacks. Results here indicate that these feedbacks may be very different for differing experimental boundary conditions. The long wave feedback due to cloud amount changes is negative in all experiments, due to a very consistent decrease in high and middle cloud fractions. Conversely, cloud height feedback is positive due to a general increase in the altitude of (particularly high) cloud. Cloud height feedback is very sensitive to the choice of the convection scheme and to the change in vertical resolution. Greatest changes in the strength of the short wave cloud feedback results from modifications to the soil moisture specification and the convection scheme. The results here indicate that large differences in cloud feedback may be diagnosed from a single model, even without changes being made to the cloud parametrization. The value of the sensitivity can thus be expected to be a function not only of the physical parametrizations chosen for the model (e.g. the penetrative convection scheme), but also of the details of the manner in which the experiment was performed (e.g. SST and soil moisture specifications). The TOA radiation perturbation analysis method proves to be a powerful technique for diagnosing and understanding the physical processes responsible for the range in climate sensitivity found between the experiments.¿ 1997 American Geophysical Union