We apply a four-dimensional version of an 'optimal detection' methodology to the near-surface temperature record over the twentieth century. The four-dimensional analysis includes variation over latitude, longitude, season and time. We compare observed surface temperature change over the twentieth century with simulations of a climate model, the Hadley Centre coupled ocean/atmosphere general circulation model (HADCM2) that include the effects of greenhouse gases, sulfate aerosols, solar variability and volcanic aerosols. For each forcing we have an ensemble of four simulations to improve our estimate of the model predicted response and, in addition, the model has been run for over 1700 years with climate forcings held constant to represent natural internal variability. Our aim is to distinguish between anthropogenic and natural causes of twentieth century climate change using an analysis that gives higher weight to seasons with high signal to noise than seasons with low signal to noise. Comparison is made with an analysis using annual mean data and the advantages of the seasonally optimised approach are discussed