When pooled for extraction as a bulk sample, the DNA within morphologically unidentifiable fossil bones can, using next-generation sequencing, yield valuable taxonomic data. This method has been proposed as a means to rapidly and cost-effectively assess general ancient DNA preservation at a site, and to investigate temporal and spatial changes in biodiversity; however, several caveats have yet to be considered. We critically evaluated the bulk bone metabarcoding (BBM) method in terms of its: (i) repeatability, by quantifying sampling and technical variance through a nested experimental design containing sub-samples and replicates at several stages; (ii) accuracy, by comparing morphological and molecular family-level identifications; and (iii) overall utility, by applying the approach to two independent Holocene fossil deposits, Bat Cave (Kangaroo Island, Australia) and Finsch's Folly (Canterbury, New Zealand). For both sites, bone and bone powder sub-sampling were found to contribute significantly to variance in molecularly identified family assemblage, while the contribution of library preparation and sequencing was almost negligible. Nevertheless, total variance was small. Sampling over 80% fewer bones than was required to morphologically identify the taxonomic assemblages, we found that the families identified molecularly are a subset of the families identified morphologically and, for the most part, represent the most abundant families in the fossil record. In addition, we detected a range of extinct, extant and endangered taxa, including some that are rare in the fossil record. Given the relatively low sampling effort of the BBM approach compared with morphological approaches, these results suggest that BBM is largely consistent, accurate, sensitive, and therefore widely applicable. Furthermore, we assessed the overall benefits and caveats of the method, and suggest a workflow for palaeontologists, archaeologists, and geneticists that will help mitigate these caveats. Our results show that DNA analysis of bulk bone samples can be a universally useful tool for studying past biodiversity, when integrated with existing morphology-based approaches. Despite several limitations that remain, the BBM method offers a cost-effective and efficient way of studying fossil assemblages, offering complementary insights into evolution, extinction, and conservation.