We assess the most probable causes of late twentieth century (1960-1994) tropospheric temperature changes. Optimal detection techniques are used to compare observed spatio-temporal patterns of near-surface and tropospheric temperature change with results from experiments performed with two different versions of the Hadley Centre climate model. We detect anthropogenic forcings, particularly well-mixed greenhouse-gases, with a less certain sulfate aerosol cooling influence. More limited evidence exists for a detectable volcanic influence. Our principal results do not depend upon the choice of model. Both models, but particularly HadCM3, appear to overestimate the simulated climate response to greenhouse gases (especially at the surface) and volcanoes. This result may arise, at least in part, due to errors in the forcings (especially sulfate) and technical details of our approach, which differs from previous studies. We use corrected and uncorrected versions of the radiosonde record to assess sensitivity of our detection results to observational uncertainties. We find that previous corrections applied to the radiosonde temperature record are likely to have been sub-optimal in only taking into account temporal consistency. However, the choice of corrected or uncorrected version has no systematic effect upon our main conclusions. We show that both models are potentially internally consistent explanations of observed tropospheric temperatures.