Marine community diversity surveys require a reliable assessment to estimate ecosystem functions and their dynamics. For these, non-invasive environmental DNA (eDNA) metabarcoding is increasingly applied in zoological studies to complement or even replace traditional morphological identification methods. However, uncertainties remain about the accuracy of the diversity detected with eDNA to capture the actual diversity in the field. Here, we validate the reliability of eDNA metabarcoding in identifying metazoan biodiversity in highly dynamic marine waters of the North Sea. We analyzed biodiversity from water (eDNA) and zooplankton samples with cytochrome c oxidase subunit 1 (COI) and 18S rRNA (18S) metabarcoding at Helgoland Roads and validated the optimal molecular resolution by morphological and molecular zooplankton identification (metabarcoding) with the result of merely a few false-negative detections. eDNA and zooplankton metabarcoding resolved 354 species from all major and in total 16 metazoan phyla. This molecular genetic species inventory overlapped by 95.9% (COI) and 81.9% (18S) with published inventories of local, morphologically identified species, among them neozoa and rediscovered species. Even though half of all species were detected by both eDNA and zooplankton metabarcoding, the methods differed significantly in their detected diversity. eDNA metabarcoding performed very well in cnidarians and annelids, whereas zooplankton metabarcoding identified higher numbers of fish and malacostraca. Species assemblages significantly differed between the individual sampling events and the cumulative number of identified species increased steadily over the sampling period and did not reach saturation. About a third of the species were detected only once while a core community of 22 species was identified continuously. Our study confirms eDNA metabarcoding to be a powerful tool to identify and analyze North Sea fauna in highly dynamic waters and we recommend investing in high sampling efforts by repetitive sampling and replication using at least 0.45 μm filters to increase filtration volume.