The biodiversity of mixed-species samples of arthropods can be characterized by shotgun sequencing of bulk genomic DNA and subsequent bioinformatics assembly of mitochondrial genomes. Here, we tested the power of mitochondrial metagenomics by conducting Illumina sequencing on mixtures of >2600 individuals of leaf beetles (Chrysomelidae) from 10 communities. Patterns of species richness, community dissimilarity and biomass were assessed from matches of reads against three reference databases, including (a) a custom set of mitogenomes generated for 156 species (89% of species in the study); (b) mitogenomes obtained by the de novo assembly of sequence reads from the real-world communities; and (c) a custom set of DNA barcode (cox1-5′) sequences. Species detection against the custom-built reference genomes was very high (>90%). False presences were rare against mitogenomes but slightly higher against the barcode references. False absences were mainly due to the incompleteness of the reference databases and, thus, more prevalent in the de novo dataset. Biomass (abundance x body length) and read numbers were strongly correlated, demonstrating the potential of mitochondrial metagenomics for studies of species abundance. A phylogenetic tree from the mitogenomes showed high congruence with known relationships in Chrysomelidae. Patterns of taxonomic and phylogenetic dissimilarity between sites were highly consistent with data from morphological identifications. The power of mitochondrial metagenomics results from the possibility of rapid assembly of mitogenomes from mixtures of specimens and the use of read counts for accurate estimates of key parameters of biodiversity directly from community samples.This article is protected by copyright. All rights reserved.