Pollination is an important ecosystem service threatened by current pollinator declines, making flower planting schemes an important strategy to recover pollination function. However, ecologists rarely test the attractiveness of chosen plants to pollinators in the field. Here, we experimentally test whether plant species roles in pollination networks can be used to identify species with the most potential to recover plant-pollinator communities. Using published pollination networks, we calculated each plant’s centrality and chose five central and five peripheral plant species for introduction into replicate experimental plots. Flower visitation by pollinators was recorded in each plot and we tested the impact of introduced central and peripheral plant species on the pollinator and resident plant communities and on network structure. We found that the introduction of central plant species attracted a higher richness and abundance of pollinators than the introduction of peripheral species, and that the introduced central plant species occupied the most important network roles. The high attractiveness of central species to pollinators, however, did not negatively affect visitation to resident plant species by pollinators. We also found that the introduction of central plant species did not affect network structure, while networks with introduced peripheral species had lower centralisation and interaction evenness than networks with introduced central species. To our knowledge, this is the first time species network roles have been tested in a field experiment. Given that most restoration projects start at the plant community, being able to identify the plants with the highest potential to restore community structure and functioning should be a key goal for ecological restoration.