The ability to identify the constituent plant species that make up a mixed-species sample of pollen has important applications in ecology, conservation, agriculture, and other areas. Recently, metabarcoding protocols have been developed for pollen, which reveal the presence/absence of plant species, but metabarcoding does not allow reliable quantification. A PCR-free, shotgun metagenomics approach has greater potential for providing reliable information on species relative abundances, but applying shotgun metagenomics to eukaryotes is challenging due to the dearth of reference genomes. We have developed a pipeline, RevMet (Reverse Metagenomics), that allows reliable and semi-quantitative characterization of the species composition of mixed-species eukaryote samples, such as bee-collected pollen, without requiring reference genomes. Instead, reference species are represented only by ‘genome skims’: low-cost, low-coverage, short-read datasets. The skims are mapped to individual long reads sequenced from mixed-species samples using the MinION, a portable nanopore DNA-sequencing device, and the long reads are uniquely assigned to plant species. We skimmed 49 wild UK plant species and used them to identify species in mock and bee-collected pollen samples, revealing plant species compositions and also differentiating high- from low-biomass species. The RevMet pipeline can be adapted to a wide range of communities of eukaryotic species.