In energy terms primary production is the driving step of the global carbon cycle. To predict the interaction of ecosystems with the "greenhoude" effect, it is necessary to understand how primary production, consumption, and decomposition will respond to climate change. Most estimates of primary production have been made by extrapolation from measured standing crops. For grasslands we show this approach to be seriously in error. Even where detailed studies of turnover and belowground production have been undertaken, errors are invariably high, severely limiting the value of models based on correlation of climate with measured production. Detailed information is available on the responses of individual plant processes to individual climatic variables at the leaf, plant, and stand level, giving potential for a more mechanistic approach in modelling. This approach is limited by lack of information on multivariate interactions and on some key physiological processes, and by uncertainties in scaling up to populations and communities. Despite this, some important insights to possible community responses, particularly those of C3 and C4 types, may be gained from knowledge of responses at the plant level and below. This review outlines the expected character of climate change in grasslands and coniferous forests. Knowledge of the responses of different physiological processes underlying production to individual aspects of climate change is considered, and its implications for higher levels of organization are discussed. Although feasible, mechanistic models of production compound the errors associated with individual process responses with uncertainties surrounding interaction and scaling up, and result in very large errors in any prediction of response to climate change. We conclude that there is insufficient information to predict accurately the response of primary production to climate change. The key processes for which information is inadequate and the parameters that have meaning at different scales need to be identified. Of particular promise is the approach of predicting production from light interception and conversion efficiency.