The Southern Ocean accounts for 10% of the world's oceans, but has a relatively low species richness. Some areas have been subject to commercial fishing, and the species compositions and population structures have been underestimated. As a promising tool, DNA barcoding has advanced to aid species identification and discovery. Even for experts, despite extensive taxonomic studies, the identification of fishes can be problematic for many reasons. In the present study, 43 Antarctic fish species were identified and analyzed using DNA barcoding, and the data obtained was examined. We collected and morphologically identified specimens from the Southern Ocean during 2013 and 2016. In total, 97 individuals from 43 species pertaining to 22 genera of Notothenioidei and Channichthyidae were barcoded using the COI gene, sequenced, and compared with that of other species recorded in GenBank and the Barcode of Life Data System (BOLD). Our results indicated that the mean T, G, A, and C contents were 31.9%, 18.3%, 22.2%, and 27.6%, the GC content of codon position 1 averaged 51.3%, that of position 3 averaged 44.0%, and that of position 2 averaged 42.0%. Therefore, the COI sequences of Antarctic fishes were obviously base biased. The average value of transition/transversion ratios was 2.79. Using the Kimera-2-parameter model, the mean genetic distance among pairwise species (0.157) was calculated as 79 times higher than that within species (0.002). From the phylogenetic tree, 41 species, except , were clustered, which suggested that COI barcoding could effectively be used to identify the remaining 41 species. But, COI barcoding provided limited phylogenetic resolution of , since they shared COI haplotypes. The similarity in results indicated that we should develop multigene barcode technology to overcome the deficiency of the single COI gene sequence. For the two species, , that were difficult to distinguish using DNA barcoding, we need more samples to analyze their phylogenetic relationships, including morphological, cytological, and molecular evolution. Although this study compared the results with that of other studies, we found that different methods and sampling could lead to different results. To determine more comprehensive phylogenetic relationships of fishes, we need to analyze a variety of methods together to form more consistent results. During the course of the study, we found that for some sister species or recently differentiated subspecies, COI barcoding mostly can not be separated, and mitochondrial and nuclear genes should be analyzed together.