This paper discusses the principles and approach required for developing small-scale management units to take account of predators’ needs when managing the fishery for Antarctic krill. It provides a theoretical foundation for considering the scales of management units involving the integration of local populations of harvested species, foraging areas of predators, fishing grounds and the potential influences of the environment, including oceanography and metapopulation structure of the harvested species. The integration of these components involves two different types of unit: the ‘harvesting unit’, which is at the scale of the metapopulation of the harvested species, and the ‘predator unit’, which does not have to be a relatively self-contained ecosystem but should be sufficiently independent for fishing in that unit not to inadvertently affect predators being monitored in other units. The South Atlantic region (Area 48) is used to illustrate how to define predator units. The derived conceptual model is then used to formulate an approach for developing fisheries on prey species, notably krill, in other harvesting units. The manner in which predator units can be used to help the CCAMLR Ecosystem Monitoring Program (CEMP) provide strategic advice on the effects of fishing is discussed. In general, the early acquisition of information on the distribution of local populations of krill and the potential foraging densities of predators from within a harvesting unit (i.e. abundance of predators, distribution of colonies and foraging range) will provide a means of circumscribing predator units as well as undertaking an assessment of long-term annual yield. It is proposed that the early development of the fishery could be concentrated in a small number of units in such a way that the relative fishing intensities in those units are equivalent, although not necessarily equal, to the intensity expected across all units once the catch limit had been reached. Other units in which fishing was not occurring could be monitored as well. This process could help determine whether or not the catch limit is likely to cause undesirable effects on predators in any of the predator units. In this way, it is possible to determine, well in advance of the catch limit being reached, whether or not local restrictions on harvesting are necessary, as well as the overall requirements for the monitoring program.