Fish play vital roles in river ecosystems; however, traditional investigations of fish usually cause certain ecological damage. Extracting DNA from aquatic environments and identifying DNA sequences offers an alternative, non-invasive approach for detecting fish species. In this study, environmental DNA (eDNA), coupled with PCR and next-generation sequencing, and electrofishing were used to compare their effects in identifying fish community characteristics. In three subtropical rivers of southern China, fish specimens and eDNA in water were collected from headwaters to estuaries. Both eDNA OTU richness and individual abundance (including number and biomass) could group 38 sampling sites into eight spatial zones with significant differences in local fish community composition. Compared with the order-/family-level grouping, the genus-/species-level grouping could more accurately recognize the differences between upstream zones I − III, midstream zones IV − V, and downstream zones VI – VIII. From headwaters to estuary, two environmental gradients significantly influenced the longitudinal distribution of fish species, including the first gradient composed of habitat and physical water parameters and the second gradient composed of chemical water parameters. The high regression coefficient of alpha diversity between eDNA and electrofishing methods as well as the accurate recognition of dominant, alien, and biomarker species at each spatial zone indicated that eDNA could characterize fish community attributes at a similar level of traditional approach. Generally, our results demonstrated that eDNA metabarcoding can be used as an effective tool in revealing fish composition and diversity, which is important for using the eDNA technique in aquatic field monitoring.