The current paper employs an advanced one equation sub-grid scale turbulence kinetic energy LES model (1-Eqn-SGS-TKE-LES) to study the design characteristics, turbulence characteristics, and flow features arising from variations in impeller design configuration in a multiphase liquid-liquid pump-mixer unit. The geometric configurations of the pump-mixer studied are: (a) an R320 curved blade impeller, with variable clearance from the tank bottom, and (b) a multiple impeller set-up produced by the addition of an upper A310 impeller. The performance of these impeller configurations have not yet been reported in the literature even though they are widely used in the solvent-extraction industry. To facilitate more effective design of these pump-mixers, improved understanding of their hydrodynamics and flow instabilities is necessary. To address this, an advanced turbulence model, the 1-Eqn-SGS-TKE-LES model, is tested in this work for single and multiphase liquid-liquid systems using both OpenFOAM and CFX-11. The predictive performance of the model for single phase operation of a pump-mixer has been evaluated by comparing with a PIV experimental dataset, the RANS k-epsilon turbulence model, the LES Smagorinsky and the LES Dynamic model. Owing to its good performance, the 1-Eqn-SGS-TKE-LES model is then extended to study the multiphase liquid-liquid system, where experimental torque measurement has been used for the model validation. The model has been used to study variation in design parameters (power number, head number, flow number), instantaneous flow structures the turbulence parameters (turbulence eddy dissipation and turbulent kinetic energy) arising out of variations in the geometric configuration of the pump-mixer. Such results are helpful in selecting a suitable design configuration. Further, this model also provides good additional information on SGS-turbulent kinetic energy. Additional improvements in the performance of the model can be expected: (a) if the effect of other sub-grid scale phenomena can be accounted for, e.g. by utilizing the additional sub-grid scale turbulent kinetic energy information to model sub-grid scale turbulent dispersion force, and (b) by enhancing the resolution of the LES model by using a much finer mesh, though this would excessively increase the computational resources required. (C) 2012 Elsevier Ltd. All rights reserved.