The biome model of Prentice et al. (1992a) is used to predict global patterns of potential natural plant formations, or biomes, from climatologies simulated by ECHAM, a model used for climate simulations at the Max-Planck-Institut fur Meteorologie. This study is undertaken in order to show the advantage of this biome model in diagnosing the performance of a climate model and assessing effects of past and future climate changes predicted by a climate model. Good overall agreement is found between global patterns of biomes computed from observed and simulated data of present climate. But there are also major discrepancies indicated by a difference in biomes in Australia, in the Kalahari Desert, and in the Middle West of North America. These discrepancies can be traced back to failures in simulated rainfall as well as summer or winter temperatures. Global patterns of biomes computed from an ice age simulation reveal that North America, Europe, and Siberia should have been covered largely by tundra and taiga, whereas only small differences are seen for the tropical rain forests. A potential northeast shift of biomes is expected from a simulation with enhanced C02 concentration according to the IPCC Scenario A. Little change is seen in the tropical rain forest and the Sahara. Since the biome model used is not capable of predicting changes in vegetation patterns due to a rapid climate change, the latter simulation has to be taken as a prediction of changes in conditions favourable for the existence of certain biomes, not as a prediction of a future distribution of biomes.[/ab]