The Hurley Pasture Model was used to examine the short and long-term responses of grazed grasslands in the British uplands to a step increase from 350 to 700 μmol mol–1 CO2 concentration ([CO2]) with inputs of 5 or 100 kg N ha–1 y–1. In N-rich grassland, [CO2] doubling quickly increased net primary productivity (NPP), total carbon (Csys) and plant biomass by about 30%. By contrast, the N-poor grassland underwent a prolonged ‘transient’, when there was little response, but eventually NPP, Csys and plant biomass more than doubled. The ‘transient’ was due to N immobilization and severe depletion of the soil mineral N pool. The large long-term response was due to slow N accumulation, as a result of decreased leaching, decreased gaseous N losses and increased N2-fixation, which amplified the CO2 response much more in the N-poor than in the N-rich grassland. It was concluded that (i) ecosystems use extra carbon fixed at high [CO2] to acquire and retain nutrients, supporting the contention of Gifford et al. (1996), (ii) in the long term, and perhaps on the real timescale of increasing [CO2], the response (in NPP, Csys and plant biomass) of nutrient-poor ecosystems may be proportionately greater than that of nutrient-rich ones, (iii) short-term experiments on nutrient-poor ecosystems may observe only the transient responses, (iv) the speed of ecosystem responses may be limited by the rate of nutrient accumulation rather than by internal rate constants, and (v) ecosystem models must represent processes affecting nutrient acquisition and retention to be able to simulate likely real-world CO2 responses.