To grow, plants need both carbon, which is fixed in photosynthesis, and inorganic nutrients, which are generally obtained from the soil. Much interest currently exists in trying to understand the uptake and storage of carbon by terrestrial ecosystems. This paper investigates to what extent carbon gain and storage are modified by soil nutrient availability. This issue is investigated in relation to both short-term carbon fluxes on the time scale of interannual variability and long-term ecosystem carbon stocks on time scales of several thousand years. We conclude from simulations with an ecosystem model (CenW) that interannual variations in carbon gain can be significantly affected by feedback effects through the nutrient cycle. This feedback effect operates principally through an imbalance between carbon and nutrient dynamics. In years that allow high carbon gain, nutrient supply typically does not match the increased carbon supply so that foliar nutrient concentrations are reduced. This lowers productivity below that which could be expected if foliar nutrient concentration remained the same. The importance of these feedback effects is shown to be greatest at intermediate levels of water availability and nutrient supply, and is relatively more important for net ecosystem carbon exchange than for net primary production. We conclude that the long-term build-up of carbon stocks in ecosystems is often controlled by the rate at which nutrients can be gained. This conclusion is based on data from published studies showing that the slow build-up of carbon matches the gain in nitrogen, phosphorus and sulfur, and on our simulations of system carbon stocks in response to fertiliser addition. The paper concludes with a discussion of the importance and feasibility of including these processes into models at different scales, including the broad continental scale. For modelling net ecosystem exchange for Australia, it is regarded as feasible and desirable to use models that are constrained by these system-internal feedback effects. Such models have already been used for large-scale simulations in Australia and other countries.