Time-slice or snapshot climate change experiments, in which an atmospheric GCM is forced by SSTs derived from different periods of a transient climate scenario, have been widely used over recent decades. This technique is much cheaper than the use of a coupled ocean-atmosphere GCM and is therefore very useful to test the sensitivity of the climate response to various changes in the lower boundary conditions (correction of SST biases, additionnal impact of land-use change) or in the horizontal resolution. Nevertheless,there is no real consensus about the design of such experiments and it is not so clear that the removal of the ocean feedback has no major influence on the simulated climate. In the present study, three pairs of time-slice experiments are performed, based on two periods (1950-1999 and 050-2099) of a transient simulation with the CNRM coupled GCM. In the first pair, the CNRM AGCM is forced directly by the climatological monthly mean SSTs derived from the coupled integration. In the second pair, the present-day SST biases are removed from the climatological SST forcings. In the third and last pair, the SST biases are also removed, but the forcing includes the interannual variability of the SST based on available observations.The results suggest that the use of climatological SSTs does not allow the AGCM to reproduce accurately the response of the coupled model to enhanced concentrations of greenhouse gases. Removing the SST biases has a significant impact on the results, as well as the use of time-evolving instead of climatological SST boundary conditions. Nevertheless, the ocean-atmosphere feedback appears as a key mechanism at the intraseasonal and interannual timescales, that cannot be neglected without influence on the predicted climate change.