Studies of grassland communities have demonstrated that increasing vertebrate grazing decreases the diversity of specialised herbivorous insects, while plant diversity is maintained or increased. However, we still have a limited understanding of the causal mechanisms underlying these contrasting observations of two tightly linked groups of organisms. We used spatially linked plant and moth observations from salt marshes, sampled for 3 years along an experimental sheep-grazing gradient (0, 1–2, 3–4 and 10 sheep ha ⁻¹), to test whether the disruption of plant–insect interactions by large herbivores accounts for these contrasting grazing effects. Moths were caught using emergence traps, which were moved and repositioned every 3 weeks. Firstly, we quantified species turnover between the grazing regimes for both taxa (measured as Sørensen dissimilarity) using a null-model approach. Secondly, we analysed the number of observed insect ̶ host associations under the different regimes. Species turnover between grazing regimes was significant (after correcting for rarefaction effects) for moth species, but not for plants, indicating very few and random effects of grazing on plant species composition. The percentage of realised plant–moth associations decreased from 37% in the absence of grazing to 6.5% under high stocking densities. We thus conclude that vertebrate grazing caused a disruption of plant–moth associations, probably by rendering the host-plants unsuitable for most of the moth species. Our findings provide further mechanistic understanding on how large herbivores shape arthropod communities and illustrate the importance of host-plant associations in explaining effects of natural or anthropogenic habitat modification.