Computational fluid dynamics (CFD) methods are frequently applied to model mechanically stirred tanks. However, if this approach is to provide a reliable design tool, accurate predictions are essential for all important characteristics. Past validation studies have generally demonstrated that mean velocities agreed reasonably well with experimental measurements, but predictions of the turbulence kinetic energy, k, and its rate of dissipation, epsilon, have been considerably poorer. However, these parameters are also important. For example, many multiphase flow phenomena are modelled as functions of k and epsilon. CFD modelling methods were investigated in relation to a tank stirred by a Lightnin A310 hydrofoil, following a configuration for which experimental data were available for mean velocities and turbulence quantities. Initial modelling based a mesh of similar to 3 million nodes showed that the Shear Stress Transport (SST) turbulence model was superior to k-epsilon for predicting flow structures, but turbulence was underpredicted. Further model development was undertaken, in which mesh resolution was increased and additional turbulence models were compared. To minimise simulation times, a reduced domain with just one impeller blade was adopted, but improvements were confirmed by running a final full-tank simulation. With a final mesh of similar to 28 million nodes and using a modified SST model, turbulence predictions were substantially improved, and integration of a over the tank volume yielded about 90% of the expected power input. An alternative Detached Eddy Simulation method was also tested, based on a reduced geometry representing one third of the tank with a mesh of similar to 13 million nodes. However, in this case integration of the energy dissipation rate yielded only 70% of the power input, suggesting that a further increase in mesh density may be required. Crown Copyright (C) 2017 Published by Elsevier Ltd. All rights reserved.