Data on fish bycatch in the krill from Commercial catch data (95 513 hauls) and CCAMLR SISO data (11875 hauls) was used to examine the frequency of occurrence, proportion by mass, length-frequency distribution and geographic provenance of the key fish taxa reported. There was a high degree of overlap in the most frequently reported taxa in the C1 data and SISO data, with Lepidonothen larseni and Chaenodraco wilsoni the most frequently reported in both datasets. The estimated total annual mass of fish by-catch in a 300 000 t krill fishery would be 370 t, comprising 40% Champsocephalus gunnari and 30% Lepidonotothen larseni. The length-frequency distribution of all taxa for which >100 fish were measured had modal size class of <10 cm. The fish species taken as by-catch in the krill fishery are the same species (and size classes) as those reported in the diet of ‘krill-dependent’ predators. There is evidence of both an increase in the data quality from the observer scheme, as reflected in the reduced confidence intervals around the frequency of occurrence data, as well as an increase in the fish by-catch reported in the commercial krill catch data. A systematic method for observers to record fish by-catch and increases in observer coverage in the krill fishery has provided an increase in the data available with which to examine the magnitude of fish by-catch in the krill fishery.