Flower-derived eDNA holds great promise as a rapid and non-invasive tool for monitoring pollinators and their plant-associations. However, pollinators often only briefly interact with a plant and leave little eDNA, making them particularly challenging to detect. In addition, taxonomic biases in eDNA deposition and PCR amplification prevent quantitative analysis of pollinator diversity. These limitations have so far precluded the widespread use of eDNA in pollinator monitoring. Comparing flower-derived eDNA with conventional monitoring in flower strips, we here explore the utility of eDNA to detect community diversity, species abundances and ecological specificity of plant-associated arthropods. We show that read abundances are a bad predictor of true abundances at the community level. Instead, the occupancy of individual species in replicated flower eDNA samples provides reliable quantitative estimates of pollinator biodiversity and detects their ecological specificity very well. Also, we find that pollinator eDNA can be collected non-invasively, by washing off from flowers in the field. Our work highlights eDNA analysis as a powerful tool for the rapid future monitoring of plant-arthropod interactions and plant-pollinator networks.