A simplified vegetation distribution prediction scheme is used in combination with the Biosphere-Atmosphere Transfer Scheme (BATS) and coupled to a version of the NCAR Community Climate Model (CCM1) which includes a mixed-layer ocean. Employed in an off-line mode as a diagnostic tool, the scheme predicts a slightly darker and slightly rougher continental surface than when BATS' prescribed vegetation classes are used. The impact of tropical deforestation on regional climates, and hence on diagnosed vegetation, differs between South America and S.E. Asia. In the Amazon, the climatic effects of removing all the tropical forest are so marked that in only one of the 18 deforested grid elements could the new climate sustain tropical forest vegetation whereas in S.E. Asia in seven of the 9 deforested elements the climate could continue to support tropical forest. Following these off-line tests, the simple vegetation scheme has been coupled to the GCM as an interactive (or two-way) submodel for a test integration lasting 5.6 yr. It is found to be a stable component of the global climate system, producing only ~ 3% (absolute) interannual changes in the predicted percentages of continental vegetation, together with globally-averaged continental temperature increases of up to + 1.5 °C and evaporation increases of 0 to 5 W m-2 and no discernible trends over the 67 months of integration. On the other hand, this interactive land biosphere causes regional-scale temperature differences of ± 10 °C and commensurate disturbances in other climatic parameters. Tuning, similar to the q-flux schemes used for ocean models, could improve the simulation of the present-day surface climate but, in the longer term, it will be important to focus on predicting the characteristics of the continental surface rather than simple vegetation classes. The coupling scheme will also have to allow for vegetation responses occurring over longer timescales so that the coupled system is buffered from sudden shocks.