Under the assumption that observed near-surface temperatures can be expressed as a linear combination of a set of externally-forced signals (both natural and anthropogenic) plus internally-generated natural variability, removal of any known signal or combination of known signals from the observed temperature record should make the residual signal(s) easier to detect. If all externally-forced signals could be removed, the residual should just be internally-generated variability: or, more generally, the residual must be the sum of this internal variability, neglected signals, and errors in the signals removed. The present analysis considers the effects of sequentially removing different combinations of CO2, sulfate aerosol, ozone and solar signals, with different relative weights chosen to reflect uncertainties in their contributions to past radiative forcing changes. This leads to a range of residuals which are examined relative to unforced, control-run O/AGCM simulations in terms of mean, variability and auto- correlation properties. The results support the hypothesis that all four external forcings are important in explaining past patterns of near-surface temperature change