We compare several Atmosphere/Ocean General Circulation model control runs by fitting long control runs to a simple discrete (annual averages) stochastic model: Rate of change of global heat content = difference of absorbed and emitted radiation + white noise. If the heat content is taken to be proportional to the surface temperature we can estimate the effective (broad frequency‐band averaged) heat capacity for the system by a regression procedure. We find a wide range of effective heat capacities that correlate significantly with the published sensitivities of the models. Models with large sensitivities have large thermal inertia. Several interpretations are discussed.