We present a direct numerical simulation (DNS) of the turbulent flow in a baffled tank driven by by a Rushton turbine. The DNS is compared to a Large Eddy Simulation (LES), a Reynolds Averaged Navier-Stokes (RANS) simulation, Laser Doppler Velocimetry data, and Particle Image Velocimetry data from the literature. By Reynolds averaging the DNS-data, we validate the turbulent viscosity hypothesis by demonstrating strong alignment between the Reynolds stress and the mean strain rate. Although the turbulent viscosity ?T in the DNS is larger than in the RANS simulation, the turbulent viscosity parameter C mu = ?T?/k2, is an order of magnitude smaller than the standard 0.09 value of the k-? model. By filtering the DNS-data, we show that the Smagorinsky constant CS is uniformly distributed over the tank with CS similar to 0.1. Consequently, the dynamic Smagorisnky model does not improve the accuracy of the LES. (C) 2012 American Institute of Chemical Engineers AIChE J, 2012