A method is proposed to diagnose climate feedbacks of water vapor, temperature lapse-rate, and cloud variations in atmospheric general circulation models. It is then applied to study differences in sensitivity of the National Center for Atmospheric Research community climate model (CCM2) and two hybrid versions of CCM2 with different cumulus-convection schemes. Water vapor feedback and temperature lapse-rate feedback differ among the models due to different efficiencies of heat and moisture transport by cumulus convections. A large compensation occurs between water vapor feedback and temperature lapse-rate feedback. This leads to similar clear-sky sensitivities in the models. Cloud-radiative feedback is negative in CCM2 with a ΔSST climate change due to the vigorous cumulus-convective scheme. Stronger convection warms the upper troposphere and reduces its cloudiness more, resulting in negative longwave cloud-radiative feedback. In models where a moist-adiabatic-adjustment scheme and then a decoupling of the atmospheric boundary layer are subsequently used, intensity of cumulus convection is successively reduced and cloud-radiative feedback changes to either neutral or positive.