Abstract: Abstract  This study used environmental DNA metabarcoding and traditional netting methods to detect the fish community structure and diversity in Dongxixi Watershed in Xiamen from 2023 to 2024 respectively in the flood season, dry season, and normal water season. The result showed that a total of 134 fish species were detected by eDNA metabarcoding, during the flood season, which was significantly higher than that in dry season and normal water season (61 and 48 species, respectively). The fish composition in different hydrological periods was dominated by Cypriniformes and Perciformes, accounting for 59.96%-99.56% of the total detected fish species. There was an ecological invasion phenomenon of Tilapias species such as Coptodon zillii, Oreochromis aureus, Oreochromis niloticus. There was no significant difference in fish alpha diversity among different sampling points in Dongxixi Watersheds. The Shannon index and Margalef index in wet season were significantly higher than those in dry season and normal season (P<0.05), while the Pielou index in dry season was significantly higher than those in wet season and normal season (P<0.05). Non-metric multidimensional scaling analysis (NMDS) shows that the heterogeneity between dry season and normal season was low and the species composition was similar, while the fish species composition in wet season and other two hydrological periods were significantly different and the heterogeneity was high. Moreover, the number of fish species detected by eDNA method was significantly higher than that captured by traditional methods, and the ecological types of fish obtained (feeding habits, habitat and tolerance) were highly consistent. This study proves that eDNA metabarcoding technology is feasible in the fishery resources survey in the watershed, and it is better than traditional morphology investigation in the number of species detected efficiency. Combined with traditional detection, it can further increase the reliability of monitoring investigation.