We report a new approach to climate change detection and attribution using an atmospheric general circulation model (AGCM), complementary to the traditional approach using coupled ocean--atmosphere models (CGCM). Ensembles of simulations were run with an AGCM forced with the observed history of sea-surface temperature (SST) and sea-ice extent and repeated with a variety of forcing factors added incrementally. SST changes alone give a warming of only about 0.15 ¿C in annual global land surface air temperature between 1950 and 1994. Addition of changing greenhouse gases, including off-line calculations of tropospheric ozone, give a further warming of 0.15 ¿C, still 0.2 ¿C less than observed. This deficit in warming derives from the Northern Hemisphere (NH) winter half-year as the summer half-year NH temperature is well-simulated. In the lower stratosphere, little cooling is simulated using the observed changes of SST alone but increasing the concentration of greenhouse gases and decreasing the concentration of stratospheric ozone leads to a cooling close to that observed. Inclusion of changes to tropospheric ozone with other forcing factors, the first time this has been attempted, gives good simulations of tropospheric and stratospheric temperature changes; these are significantly more similar to observations than using SST variations alone. Despite the uncertainties, these simulations strongly indicate a discernible anthropogenic effect on the annual mean thermal structure of the atmosphere, the first time this has been shown in the presence of the observed variations of SST and sea-ice extent. ¿ 1998 American Geophysical Union