Most animal pollination results from plant–insect interactions, but how we perceive these interactions may differ with the sampling method adopted. The two most common methods are observations of visits by pollinators to plants and observations of pollen loads carried by insects. Each method could favour the detection of different species and interactions, and pollen load observations typically reveal more interactions per individual insect than visit observations. Moreover, while observations concern plant and insect individuals, networks are frequently analysed at the level of species. Although networks constructed using visitation and pollen‐load data have occasionally been compared in relatively specialised, bee‐dominated systems, it is not known how sampling methodology will affect our perception of how species (and individuals within species) interact in a more generalist system. Here we use a Diptera‐dominated high‐Arctic plant–insect community to explore how sampling approach shapes several measures of species' interactions (focusing on specialisation), and what we can learn about how the interactions of individuals relate to those of species. We found that species degrees, interaction strengths, and species motif roles were significantly correlated across the two method‐specific versions of the network. However, absolute differences in degrees and motif roles were greater than could be explained by the greater number of interactions per individual provided by the pollen‐load data. Thus, despite the correlations between species roles in networks built using visitation and pollen‐load data, we infer that these two perspectives yield fundamentally different summaries of the ways species fit into their communities. Further, individuals' roles generally predicted the species' overall role, but high variability among individuals means that species' roles cannot be used to predict those of particular individuals. These findings emphasize the importance of adopting a dual perspective on bipartite networks, as based on the different information inherent in insect visits and pollen loads.