AbstractSpecies extinctions undermine ecosystem functioning but species do not disappear at random. Ecosystem dynamics are ruled by a subset of dominant species, but it is not clear how losing several of these key species would affect plant-pollinator interactions. We tested how the experimental loss of the most highly visited generalist plants would affect flower visitor abundances (visitation) and pollinator effectiveness (quantity of pollen-tubes within pistils) in several sites. Significant changes in the plant-pollinator communities were detected. Firstly, visitation decreased after removing 1-2 most visited plants, suggesting that these species mostly facilitate other plants by keeping high flower visitor abundances in the sites. Nevertheless, we recorded within-site variations of these trends, especially among sampled transects and among plant species, suggesting that complex facilitation-competition interactions occur among plants, but that these vary among transects within the same sites. Pollinator effectiveness fluctuated but not in a clear linear way and was not directly linked to changes of flower visitation. This suggests that fluctuations of pollinator effectiveness might be due to destabilization of the pollinators by the removal of key resources. In addition, we detected a constrained flexibility of the foraging of flower visitors because they did not switch among flower shapes and they favoured specific plant traits such as high sugar content, small inflorescences and taller plants. Indeed, these constraints would eventually limit utilisation of new resources after perturbations, possibly undermining the stability of the system. In conclusion, our work demonstrates that the loss of dominant, highly generalist plants alters plant-pollinator interactions with implications for pollination and insect foraging. Therefore, in order to preserve ecosystems, generalist plants should not get lost, because they sustain the complex pattern of interactions between plants and flower visitors.