Based on an attribution analysis of the global mean temperature biases in the Flexible Global Ocean-Atmosphere-Land System model, spectral version 2 (FGOALS-s2) through a coupled atmosphere-surface climate feedback-response analysis method (CFRAM), the model’s global surface-atmosphere energy balance in boreal winter and summer is examined. Within the energy-balance-based CFRAM system, the model temperature biases are attributed to energy perturbations resulting from model biases in individual radiative and non-radiative processes in the atmosphere and at the surface. The results show that, although the global mean surface temperature (Ts) bias is only 0.38 K in January and 1.70 K in July, and the atmospheric temperature (Ta) biases from the troposphere to the stratosphere are only around ±3 K at most, the temperature biases due to model biases in representing the individual radiative and non-radiative processes are considerably large (over ±10 K at most). Specifically, the global cold radiative Ts bias, mainly due to the overestimated surface albedo, is compensated for by the global warm non-radiative Ts bias that is mainly due to the overestimated downward surface heat fluxes. The model biases in non-radiative processes in the lower troposphere (up to 5–15 K) are relatively much larger than in upper levels, which are mainly responsible for the warm Ta biases there. In contrast, the global mean cold Ta biases in the mid-to-upper troposphere are mainly dominated by radiative processes. The warm/cold Ta biases in the lower/upper stratosphere are dominated by non-radiative processes, while the warm Ta biases in the mid-stratosphere can be attributed to the radiative ozone feedback process.