Environmental DNA (eDNA) metabarcoding is revolutionizing the study of aquatic ecosystems, enabling high-throughput analysis of biodiversity with minimal disturbance. Despite its potential to support fisheries management, species identification and downstream analysis reliability are hindered by the lack of standardization in DNA fragment choice. This study compares the species discrimination power of three markers used in marine fish eDNA metabarcoding —12S rRNA, 16S rRNA, and cytochrome oxidase subunit I (COI)— as well as two amplicons for each. We analyzed sequences from NCBI GenBank for 10 orders of Actinopterygii, including mitochondrial genomes. We assessed species discrimination by determining the percentage of monophyletic species in Neighbor-Joining trees and calculating average congeneric divergences for two data sets: one with genomic regions extracted from mitochondrial genomes (771 species) and another with independent sequences for each region (3879 species). Among (meta)barcoding amplicons, for the mitochondrial genomes’ dataset, the Folmer and Leray-Lobo regions had the highest discriminatory power, with 89.2% and 87% monophyletic species, respectively, while the 12S Teleo region had the lowest at 71.6%. Conversely, using independent sequences, the Folmer and Leray-Lobo regions had the lowest percentages of monophyletic species, at 64.8% and 63.5%, respectively, while Ac16S had the highest at 83.0%. Species discrimination is influenced by the marker’s substitution rate, fragment length, target fish order, and the quality of reference sequence data. We recommend considering species discriminatory power differences for amplicon selection, especially for species-level identifications. We advise a standard multi-marker approach under certain scenarios, namely when the presence of close congeneric species is expected.