Turbulence predictions in stirred tanks have long been underpredicting the levels of turbulence in the region near the impeller. One reason for this may be that the use of the Reynolds average approach is inadequate in this region and large eddy simulation may offer a way forward in terms of accuracy. However, the improvement in the accuracy of the simulation has to be balanced against the increase in computer resources required to calculate the answer. It would therefore be helpful if a methodology existed with which the CFD user could specify the sub-regions of the simulation where a large eddy simulation should be performed, and be able to allocate larger computational resources to it, with minimal loss of information at the interface between the two modelling methodologies. This has been achieved for vortex shedding from a bluff body, which still requires accurate modelling of vortex shedding in transient flow. A potential saving in CPU time over a complete large eddy simulation has also been demonstrated. (C) 2004 Elsevier Ltd. All rights reserved.