A Management Procedure (MP) approach is proposed to assist in advising regarding the subdivision of the precautionary catch limit for krill among 15 small-scale management units (SSMUs) in the Scotia Sea to reduce the potential impact of fishing on land-breeding predators. The Spatial Multi-species Operating Model (SMOM) developed in Plaganyi and Butterworth (2006) is used as an operating model which simulates the “true” dynamics of the resource with tests across a wide range of scenarios for the underlying dynamics of the resource. Unlike static catch allocation options, the illustrative MPs developed here have a feedback structure, and hence are able to react and self-correct. It is important, as with the static allocation options, to ensure that the likely performances of these MPs in terms of low risk to predators within each SSMU are reasonably robust to the primary uncertainties about such dynamics. A MP module separate from the operating model contains the methods and rules that are used to subdivide the krill catch between SSMUs. Different MPs are then simulation tested with their performances being evaluated on the basis of a set of performance statistics which essentially compare the risks of reducing the abundances of predators (and krill) below certain levels, as well as the variability in future average krill catches per SSMU associated with each MP. The key assumption made here is that data will be regularly available in future to monitor the impact/s of different krill catch limits. For illustrative purposes, it is assumed that two main sources of data will be available for use in a MP: (1) indices of absolute or relative abundance, or performance of the various predators (i.e. the CEMP series), and (2) survey estimates of krill absolute or relative abundance per SSMU. The approach proposed is readily modified if, for example, no krill abundance indices are available. Given that “future” data are required as inputs to test a MP including feedback, these data are generated with random variation about their underlying values and assuming the same variance as estimated from the past data.