Current conceptual models predict that an increase in stress shifts interactions between plants from competitive to facilitative; hence, facilitation is expected to gain in ecological importance with increasing stress. Little is known about how facilitative interactions between plants change with increasing biotic stress, such as that incurred by consumer pressure or herbivory (i.e. disturbance sensu Grime). In grazed ecosystems, the presence of unpalatable plants is reported to protect tree saplings against cattle grazing and enhance tree establishment. In accordance with current conceptual facilitation-stress models, we hypothesised a positive relationship between facilitation and grazing pressure. We tested this hypothesis in a Weld experiment in which tree saplings of four diVerent species (deciduous Fagus sylvatica, Acer pseudoplatanus and coniferous Abies alba, Picea abies) were planted either inside or outside of the canopy of the spiny nurse shrub Rosa rubiginosa in enclosures diVering in grazing pressure (low and high) and in exclosures. During one grazing season we followed the survival of the diVerent tree saplings and the level of browsing on these; we also estimated browsing damage to the nurse shrubs. Shrub damage was highest at the higher grazing pressure. Correspondingly, browsing increased and survival decreased in saplings located inside the canopy of the shrubs at the high grazing pressure compared to the low grazing pressure. Saplings of both deciduous species showed a higher survival than the evergreens, while sapling browsing did not diVer between species. The relative facilitation of sapling browsing and sapling survival – i.e. the diVerence between saplings inside and outside the shrub canopy – decreased at high grazing pressure as the facilitative species became less protective. Interestingly, these Wndings do not agree with current conceptual facilitationstress models predicting increasing facilitation with abiotic stress. We used our results to design a conceptual model of facilitation along a biotic environmental gradient. Empirical studies are needed to test the applicability of this model. In conclusion, we suggest that current conceptual facilitation models should at least consider the possibility of decreasing facilitation at high levels of stress.