1. A common structural feature of natural communities is the non-random distribution of pairwise interactions between organisms of different trophic levels. For plant–animal interactions, it is predicted that both stochastic processes and functional plant traits that facilitate or prevent interactions are responsible for these patterns. 2. However, unbiased manipulative field experiments that rigorously test the effects of individual traits on community structure are lacking. We address this gap by manipulating floral scent bouquets in the field. 3. Manipulation of floral scent bouquets led to quantitative as well as qualitative restructuring of flower–visitor networks, making them more generalized. Olfactometer trials confirmed both positive and negative responses to scent bouquets. 5. Our results clearly show that the distribution of insect visitors to the two abundant study plant species reflects the insects' species-specific preferences for floral scents, rather than for visual or morphological floral traits. 5. Thus, floral scents may be of major importance in partitioning flower–visitor interactions. Integrating experimental manipulations of plant traits with field observations of interaction patterns thus represents a promising approach for revealing the processes that structure species assemblages in natural communities.