High-thermal performance PWR (pressurized water reactor) spacer grids require both low pressure loss, high wall heat transfer coefficient and high critical heat flux (CHF) properties. A further detailed understanding of the main physical phenomena (wall boiling, entrainment of bubbles in the wakes, recondensation) is needed and can be reached by numerical simulation. In the field of fuel assembly analysis or design by means of CFD codes, the overwhelming majority of the studies are carried out using two-equation eddy viscosity models (EVM) as turbulence model, especially the standard K − ɛ model, while the use of Reynolds Stress Transport Models (RSTMs) remains exceptional. In contrast, extensive testing and application over the past three decades have revealed a number of shortcomings and deficiencies in eddy viscosity models. Indeed, the K − ɛ model is totally blind to flow rotation, e.g., in the presence of swirls. This aspect is crucial for the simulation of a hot channel in a fuel assembly. In fact, the mixing vanes of the spacer grids generate swirls in the coolant water, which enhances the heat transfer from the rods to the coolant in the hot channels and thus limits boiling. In this work, all the models are implemented in NEPTUNE_CFD, a three dimensional multi-fluid code developed especially for nuclear reactor applications, in the framework of the NEPTUNE project (EDF, CEA, AREVA-NP, IRSN). In a previous work, we have evaluated computational fluid dynamics results obtained with RSTM against single-phase liquid water tests equipped with a mixing vane and against two-phase boiling cases (DEBORA-tube and ASU-annular channel tests). In the present work, a geometry closer to actual fuel assemblies is considered. It consists of a rectangular test section in which a 2 × 2 rod bundle, equipped with a simple spacer grid with mixing vanes, is inserted. The influence of the turbulence model on target variables supposed to be related to CHF limitation is discussed. Numerical investigations on the optimised angle of mixing vanes are addressed. In addition, the impact of a detailed description of the bubble size distribution on the boiling flow with realistic geometry is addressed. The relative importance of the correct description of this modelling of polydisperse bubble size is due to the dependency of (i) main hydrodynamic forces, like drag, as well as of (ii) inter-phase transfer area with respect to the individual bubble size. The effects of the phenomena of coalescence or break-up and the main influences of mean bubble diameter on the void fraction and fluid velocity in the vicinity of the grids are underlined. The study of this 2 × 2 rod bundle case is a further step towards a physically reliable local CFD modelling of the two-phase boiling flow in real fuel assemblies, including spacers grid structures.