Environmental DNA metabarcoding (eDNA metabarcoding) is a fast, efficient and environmentally friendly method to investigate the diversity of fish species, but its sensitivity and reliability remain to be further evaluated. In this study, Gaobi Stream and Chengping Reservoir, located in the source region of Songyin Stream, a first level tributary of Oujiang River, were selected as the investigated waters, and the survey on fish species distributed in those two waters was carried out by traditional capture-based method and eDNA metabarcoding technology, respectively. The results showed that a total of 24 species of fish were found in the two waters by capture-based method, 88% (21 species) of fish species were recorded in the Oujiang River system. 12 fish species belonging to 4 orders, 8 families, and 11 genera were collected from Gaobi River, and 16 species of fish belonging to 3 orders, 4 families and 14 genera were collected from Chengping Reservoir. A total of 31 species were detected in the two waters by eDNA metabarcoding technology, and 84% (26 species) of fish species were recorded in the Oujiang River system. 28 fish species were found in Gaobi Stream, belonging to 5 orders, 12 families and 25 genera, and 27 species were found in Chengping Reservoir, belonging to 6 orders, 11 families, and 25 genera. However, the two survey methods found only one fish species together in the stream water and only five in the reservoir water. In addition, Shannon diversity index, Simpson diversity index, Margalef’s richness index and Pielou’s evenness of Gaobi Sream were higher than those of Chengping Reservoir, and the conclusions of the two survey methods were relatively consistent. In conclusion, eDNA metabarcoding can find more species and has higher sensitivity than the traditional method, but its results of fish species composition have a marked deviation from those of actual capture. Therefore, eDNA metabarcoding technology can only be used as reference for traditional fishery survey methods at present, and it is necessary to be cautious in drawing survey conclusions by using this technology.