This study investigates the response of a climate model to a gradual increase or decrease of atmospheric carbon dioxide. The model is a general circulation model of the coupled atmosphere-ocean-land surface system with global geography and seasonal variation of insulation. To offset the bias of the coupled model toward settling into an unrealistic state, the fluxes of heat and water at the ocean-atmosphere interface are adjusted by amounts that vary with season and geography but do not change from one year to the next. Starting from a quasi-equilibrium climate, three numerical time integrations of the coupled model are performed with gradually increasing, constant, and gradually decreasing concentration of atmospheric carbon dioxide. It is noted that the simulated response of sea surface temperature is very slow over the northern North Atlantic and the Circumpolar Ocean of the Southern Hemisphere where vertical mixing of water penetrates very deeply. However, in most of the Northern Hemisphere and low latitudes of the Southern Hemisphere, the distribution of the change in surface air temperature of the model at the time of doubling (or halving) of atmospheric carbon dioxide resembles the equilibrium response of an atmospheric-mixed layer ocean model to CO2 doubling (or halving). For example, the rise of annual mean surface air temperature in response to the gradual increase of atmospheric carbon dioxide increases with latitudes in the Northern Hemisphere and is larger over continents than oceans. When the time-dependent response of the model oceans to the increase of atmospheric carbon dioxide is compared with the corresponding response to the CO2, reduction at an identical rate, the penetration of the cold anomaly in the latter case is significantly deeper than that of the warm anomaly in the former case. The lack of symmetry in the penetration depth of a thermal anomaly between the two cases is associated with the difference in static stability, which is due mainly to the change in the vertical distribution of salinity in high latitudes and temperature changes in middle and low latitudes. Despite the difference in penetration depth and accordingly, the effective thermal inertia of the oceans between the two experiments, the time-dependent response of the global mean surface air temperature in the CO2 reduction experiment is similar in magnitude to the corresponding response in the CO2 growth experiment. In the former experiment with a colder climate, snow and sea ice with high surface albedo cover a much larger area, thereby enhancing their positive feedback effect upon surface air temperature. On the other hand, surface cooling is reduced due to the larger effective thermal inertia of the oceans. Because of the compensation between these two effects, the magnitude of surface air temperature response turned out to be similar between the two experiments.