1. The aim of this study was to identify whether plant species show consistent responses to livestock grazing. The analyses were based on 35 published studies from Australian rangelands providing 55 species response lists. The primary data set comprised 1554 responses from 829 species. 2. Eight-hundred and twenty-nine species were categorized as increasers, decreasers or neutral under grazing. Of 324 species that occurred in at least two response lists, 133 (41%) responded inconsistently, increasing at least once and decreasing at least once. While 59% of species responded consistently, these results suggest that our ability to predict vegetation change under grazing is limited. 3. Particular species were not inherently more or less consistent. Rather, as species occurred in more trials, the likelihood of at least one opposite response increased; no species that occurred at least eight times was wholly consistent. A binomial model indicated that the probability of an opposite response, across all species, was 0.275. 4. Contrary responses within species must result from context rather than from species' traits. Species were more likely to decrease in response to grazing at lower rainfall than at higher rainfall. Forbs tended to increase under grazing at sites where wet seasons were cooler. Changing the grazing animal was weakly correlated with change in response direction, although not enough for it to be useful for manipulating pasture composition. We found little support for ideas that different responses within species are due to differences in alternative forage available, or due to non-linearity of response to grazing intensity. 5. At present it appears we can predict species response direction about three-quarters of the time, at a continental scale. This represents an upper limit of the reliability of prediction based on species' traits alone. Presently we do not know what aspects of the context might allow us to predict reliably the remaining one-quarter of responses.