Our limited understanding of terrestrial ecosystem responses to elevated CO2 is a major constraint on predicting the impacts of climate change. A change in botanical composition has been identified as a key factor in the CO2 response with profound implications for ecosystem services such as plant production and soil carbon storage. In temperate grasslands, there is a strong consensus that elevated CO2 will result in a greater physiological stimulus to growth in legumes and to a lesser extent forbs, compared with C3 grasses, and the presumption this will lead in turn to a greater proportion of these functional groups in the plant community. However, this view is based on data mainly collected in experiments of three or less years in duration and not in experiments where defoliation has been by grazing animals. Grazing is, however, the most common management of grasslands and known in itself to influence botanical composition. In a long-term Free Air Carbon Dioxide Enrichment (FACE) experiment in a temperate grassland managed with grazing animals (sheep), we found the response to elevated CO2 in plant community composition in the first 5 years was consistent with the expectation of increased proportions of legumes and forbs. However, in the longer term, these differences diminished so that the proportions of grasses, legumes and forbs were the same under both ambient and elevated CO2. Analysis of vegetation before and after each grazing event showed there was a sustained disproportionately greater removal (‘apparent selection’) of legumes and forbs by the grazing animals. This bias in removal was greater under elevated CO2 than ambient CO2. This is consistent with sustained faster growth rates of legumes and forbs under elevated CO2 being countered by selective defoliation, and so leading to little difference in community composition.