In this paper, we develop an ecosystem-based, precautionary management procedure for krill fisheries which draws together past experience in CCAMLR. It provides an empirical ecosystem assessment model, a decision rule for determining local scale catch limits based on a harvest strategy and a single-species assessment of yield, and a method for implementing the procedure. The decision rule for setting catch limits for a given harvest strategy has a straight forward expression of the target conditions to be achieved and the uncertainties that need to be managed and does not assume an understanding of predator-prey dynamics beyond that evident in the data. It is a natural extension of the current precautionary approach of CCAMLR for krill and can utilise existing datasets, including B0 surveys, local scale monitoring of krill densities, local-scale monitoring of predator performance, monitoring of predator foraging locations and time series of catches from the fishery. This procedure provides a common framework for inserting data, assessment methods and candidate modelling approaches for assessing yield. Consequently, its formalism means there is no need to undertake a staged approach in providing advice. The advice can be updated as improvements are made in any component of the procedure, including the provision of data, implementation of new assessment or projection models or a revision of the decision rule. This framework formalises the decisions that need to be made in dealing with an ensemble of food web models for providing suitably precautionary advice on how to spatially structure krill fisheries to account for the needs of predators. It provides the primary expectation for managing uncertainty, either by obtaining better estimates of parameters for the projection models and/or by altering the harvest strategy. Consequently, a preferred harvest strategy, which is initially untenable because of the uncertainties associated with its ecosystem impacts, could become a suitable option if its related uncertainties are reduced. Conceivably, the procedure outlined here could be used in a spatially-structured feedback management system that can ensure CCAMLR is able to respond to trends in the status of the ecosystem, including trends arising from climate change.