The performance of several steady Reynolds-averaged Navier-Stokes turbulence models and boundary-layer modelling approaches is evaluated for a single sphere, by comparison with empirical data for a Reynolds number range of 10-3.2 x 10(4). A sphere serves here as a representative model for many spherical food products. The shear stress transport (SST) k-omega turbulence model performs exceptionally well when combined with low-Reynolds number modelling (LRNM) of the boundary layer, which confirms that the turbulence model characteristics are particularly suitable to deal with this specific flow problem. Especially the k-epsilon turbulence models are less accurate at higher Reynolds numbers (>10(2)). Boundary-layer modelling with wall functions (WFs) leads to inaccurate flow-field and scalar transfer predictions, compared to LRNM. However, LRNM grids and their inherently higher computational cost are often not practically feasible, leaving WFs as the only option. It is shown that using cell sizes on the sphere surface of a few millimetres, typical for CFD studies on food products, can compromise accuracy, and grids with smaller cell sizes are actually required. (C) 2012 Elsevier Ltd. All rights reserved.