Previous research has shown that various fluxes of carbon from and into ecosystems are correlated with summary climatic measures, such as actual evapotranspiration (AET). The best known of these is a regression of net primary production of terrestrial vegetation against AET published by Rosenzweig (1968). Rosenzweig intended this regression to represent steady state relationships of net primary production to climate. Nevertheless, it is tempting to use such regressions to predict transient responses of carbon flux to climate change, and several models take such an approach. Here, using a more detailed ecosystems model, we show that lags in population responses to climate change and non-linear changes in soil nitrogen availability that limit tree growth cause large departures from this regression during the transition between current climate and a 2 × CO2 climate. Simple models that do not consider population or soil dynamics may err when applied to the period of transition during a changing climate.