Abstract: To compare the similarities and differences in fish diversity between eDNA metabarcoding and trawl net, this study, taking the southern coastal waters of Zhejiang as a case area, analyzed the fish composition, species dominance, and community diversity index of the two survey methods using bioinformatics, traditional morphological identification, non-metric multidimensional scaling analysis, significance test of differences and linear regression analysis. The results are as follows: ① A considerable number (19.92%) of non-indigenous fish OTUs (e.g., freshwater fish, deep-sea fish, tropical coral reef fish) were detected in the high-throughput sequencing results; although the sampling water layer (surface, mid-depth, bottom) did not affect the number of fish OTUs detected by eDNA (P≥0.073), it did influence the composition of fish OTUs (stress=0.138, R=0.117, P=0.001); increasing sampling density (number of replicates, number of water layers) significantly increased the number of fish OTUs (P≤1.1×10^-3); a significant proportion (24.402%) of fish OTUs in this area were in a low annotation state, and some fish sequences exhibited interspecific identity or intraspecific variation. ② The number of fish species detected by eDNA was much higher than that by trawl net (152 species vs. 75 species), and there was a significant linear correlation between the two survey methods (P=0.049 7, r_Pearson=0.381, R²=0.153). ③ When all fish species were analyzed, the two survey methods showed a significant difference in species dominance (P=0.034); for many shared fish species and dominant/keystone dominant species, such as Engraulis japonicus and Nibea albiflora, the species dominance differed by 1-2 orders of magnitude between the two survey methods. ④ Among the 12 frequently occurring fish species, only Harpadon nehereus and Ctenotrypauchen chinensis showed a linear correlation between sequence abundance and individual counts/weight, while the r_Pearson values in the fitted models for the remaining species were low or even negative. ⑤ The community diversity index measured by the two survey methods differed considerably, and their linear correlation was also very low (P≥0.087). This study demonstrates that: ① eDNA has considerable uncertainty; ② Except for some indicators, species, or stations, it is difficult to model and predict the corresponding values of one survey method based on the results of the other; ③ The two survey methods cannot replace each other but rather have a complementary and cross-validating relationship. This study provides a basis for understanding the advantages and disadvantages of eDNA metabarcoding and for the rational use of different survey methods.