Impacts of climate change on vegetation are often summarized in biome maps, representing the potential natural vegetation class for each cell of a grid under current and changed climate. The amount of change between two biome maps is usually measured by the fraction of cells that change class, or by the kappa statistic. Neither measure takes account of varying structural and floristic dissimilarity among biomes. An attribute-based measure of dissimilarity (ΔV) between vegetation classes is therefore introduced. ΔV is based on (a) the relative importance of different plant life forms (e.g. tree, grass) in each class, and (b) a series of attributes (e.g. evergreen-deciduous, tropical-nontropical) of each life form with a weight for each attribute. ΔV is implemented here for the most used biome model, BIOME 1 (Prentice, I. C. et al., 1992). Multidimensional scaling of pairwise ΔV values verifies that the suggested importance values and attribute weights lead to a reasonable pattern of dissimilarities among biomes. Dissimilarity between two maps (ΔV) is obtained by area-weighted averaging of ΔV over the model grid. Using ΔV, present global biome distribution from climatology is compared with anomaly-based scenarios for a doubling Of atmospheric CO2 concentration (2 x CO2), and for extreme glacial and interglacial conditions. All scenarios are obtained from equilibrium simulations with an atmospheric general circulation model coupled to a mixed-layer ocean model. The 2 x CO2 simulations are the widely used OSU and GFDL runs from the 1980's, representing models with low and high climate sensitivity, respectively. The palaeoclimate simulations were made with CCM1, with sensitivity similar to GFDL. ΔV values for the comparisons of 2 x CO2 with present climate are similar to values for the comparisons of the last interglacial and mid-Holocene with present climate. However, the two simulated 2 x CO2 cases are much more like each other than they are to the simulated interglacial cases. The largest ΔV values were between the last glacial maximum and all other cases, including the present. These examples illustrate the potential of ΔV in comparing the impacts of different climate change scenarios, and the possibility of calibrating climate change impacts against a palaeoclimatic benchmark.