A global land cover classification data set is used to divide the globe into seven regions to study surface temperature changes over different vegetation/surface classes. Statistically significant warming is found from the year 1900 over all regions (except for the ice sheets over Greenland and Antarctica). Outputs from three coupled climate models (CGCM2, HadCM2 and the Parallel Climate Model (PCM)) are used to examine the detection and attribution of surface temperature trends over the various vegetation classes for the past half century. An anthropogenic warming trend is detected in six of the seven regions. Observed trends are consistent with those simulated in response to greenhouse gas and sulfate aerosol forcing except over tropical forest and water where the models appear to overestimate the warming. The similarity between the resultant scaling factors for each region from the different models underscores the reliability of our detection results. Copyright © 2006 Royal Meteorological Society.