In order to review, and possibly refine, the concept of radiative forcing as a suitable metric for climate change, the responses of three general circulation models to distinct forcing scenarios are compared. CO, solar radiation, and O are added in different locations, whilst keeping the globally averaged radiative forcing constant at 1 Wm. The three models react differently to the forcings, as feedback mechanisms such as sea-ice albedo and clouds behave differently in each model. However, we find that their climate sensitivities λ (defined as the ratio of the globally averaged surface temperature change to the radiative forcing), normalised by the climate sensitivity for a control case (e.g. CO added globally), match each other to within 30% in most experiments. Moreover, the models indicate generic deviations of λ from the case of global CO perturbations: upper tropospheric O increases generally produce lower values of λ, while lower stratospheric O perturbations lead to higher values of λ, as found in some previous work. λ tends to be higher for extratropical forcings than tropical forcings; a phenomenon which can be partially accounted for by a new explanation based on the variation of the outgoing longwave radiation with latitude. Our results suggest that if the radiative forcing associated with some perturbation is multiplied by some factor accounting for the efficiency of that mechanism, then such modified forcings can be compared more robustly than the forcings themselves.