1. Anthropogenic stressors have strong impacts on ecosystems. To understand their influence, detailed knowledge about trophic relationships among species is critical. However, this requires both exceptional resolution in dietary assessments and sampling breadth within communities, especially for highly-diverse, tropical ecosystems. 2. We used gut content metabarcoding across a broad range of coral reef fishes (8 families, 22 species) in Mo’orea, French Polynesia to test whether this technique has the potential to capture the structure of a hyperdiverse marine food web. Moreover, we explored whether taxonomic groups (families) and traditional, broad-scale trophic assignments explained fish diet across four different metrics of quantifying predator-prey interactions. 3. Metabarcoding yielded a large number (4,341) of unique operational taxonomic units (i.e. prey) with high-resolution taxonomic assignments (i.e. often to the level of genus or species). We demonstrate that across multiple metrics, taxonomic group at the family level is a consistently better, albeit still weak, predictor of empirical trophic relationships than frequently-used, broad-scale functional assignments. Our method also reveals a complex trophic network with fine-scale partitioning among species, further emphasizing the importance of examining fish diets beyond broad trophic categories. 4. We demonstrate the capacity of metabarcoding to reconstruct diverse and complex food webs with exceptional resolution, a significant advancement from traditional food web reconstruction. Further, this method allows us to pinpoint the trophic niche of species with niche-based modelling, even across hyperdiverse species assemblages such as coral reefs. In conjunction with complementary techniques such as stable isotope analysis, applying metabarcoding to whole communities will provide unparalleled information about energy and nutrient fluxes and inform their susceptibility to disturbances even in the world’s most diverse ecosystems.