Fish aquaculture potentially affects the local ecosystem through feed (leftovers) and feces. However, evaluation of its impacts on biodiversity, especially those of mobile organisms such as fish, is not straightforward. Here, we report a case study in which environmental DNA (eDNA) metabarcoding with MiFish universal primers for the 12S rRNA region of fish is used to detect and assess the impact of aquaculture on local fish communities. In Shizugawa Bay, Minamisanriku Town, Miyagi Prefecture, Japan, Oncorhynchus kisutch (coho salmon) is aquacultured in sea cages from November to July, but is absent from August to October. We conducted eDNA surveys in June, August, October, and December 2021 of water at depths of 5 m and 1 m above the bottom obtained from two sites, one approximately 100 m and the other approximately 2500 m from the nearest aquaculture site. The total number of operational taxonomic units (OTUs) detected was 122, with a maximum of 50 and a minimum of 10. With distance‐based redundancy analysis (dbRDA) and PERMANOVA applied to the eDNA‐based community compositional data, we detected a significant effect of the presence of O. kisutch aquaculture on fish biodiversity. The generalized linear model (GLM) applied to each detected OTUs indicated that the aquaculture was associated positively with Leucopsarion petersii (ice goby), and Oncorhynchus spp. and negatively with Chaenogobius annularis (forktongue goby). Environmental DNA metabarcoding surveys, combined with multivariate statics, are a promising tool to identify the dimension of biodiversity that human activity affects and would serve as a basic survey for sustainable use of the sea.