This study investigates changes to the annual temperature cycle in both observed records and output from a coupled ocean-atmosphere global climate model. Using least-squares harmonic analysis, changes to the observed annual harmonic (for the time period 1856-1998), in addition to the 1961-1990 climatology, are compared with 9 simulations from the HadCM2 model. The first simulation is a 1400 yr control integration, whilst the remainder are from 2 ensembles representing (1) increases in CO2 concentrations and (2) a combination of CO2 and sulphate aerosol increases. Observed and simulated climatologies are generally comparable, although large amplitude and phase discrepancies exist over northern North America and high-latitude oceans, respectively. The agreement may be partly artificial over the oceans due to the use of flux adjustments to maintain a realistic sea-surface temperature field. Observed northern hemisphere amplitude decreases during the 20th century agree well with simulated changes, although there are some regional differences; observed changes to the southern hemisphere amplitude are insignificant. The sign of northern hemisphere phase changes are opposite in the 2 data sets. The nature of these results is unchanged after consideration is given to the varying spatial coverage of the observed data set, by means of applying a frozen grid mask to both observed and simulated data. These findings are consistent with previous studies, though we extend them by updating the observed record, by using ensembles to better define the climate change signal, and by considering the direct effects of sulphate aerosols. For a given warming, the inclusion of aerosols results in an enhanced amplitude decrease within the northern hemisphere, related to the summertime maximum of the direct sulphate cooling effect.