Due to the difficulty of tracking microbial dispersal, it is rarely possible to disentangle the relative importance of dispersal and species sorting for microbial community assembly. Here, we leverage a detailed multilevel network to examine drivers of bacterial community assembly within flowers. We observed flower visitors to 20 focal plant species in a coflowering community in the Sierra Nevada, revealing 289 species of arthropods. We also analyzed bacterial communities on flowers of each species. We found that plant species with similar visitor communities tend to have similar bacterial communities, and visitor identity to be more important than plant relatedness in structuring floral bacterial communities. However, plant species that were hubs of arthropod visitation were not necessarily hubs of floral bacteria, suggesting an important role for species sorting. Across plant species, the composition of flower-visiting Diptera (flies), bees and non-bee Hymenoptera best predicted bacterial species composition on flowers. Taken together, our analyses suggest dispersal is important in determining similarity in microbial communities across plant species, but not as important in determining the overall macrostructure (nestedness, modularity) and microstructure (connectedness based on shared interactors) of the floral bacterial network. A multilevel network approach thus allows us to address features of community assembly that cannot be considered when viewing networks as separate entities.