In-cylinder large scale and small scale structures are widely recognized to strongly influence the mixing process in HSDI Diesel engines, and therefore combustion and pollutant emissions. In particular, swirl motion intensity and temporal evolution during the intake and compression strokes must be correctly estimated to properly target the spray jets. The experimental characterization of the attitude of a valve/port assembly to promote swirl is traditionally limited to the steady flow bench, in which the analysis is carried out for fixed valve positions / fixed pressure drops and with no piston. Since flow bench analyses cannot reproduce the highly complex instantaneous flow conditions typical of actual engine operations, the use of fully-transient in-cylinder numerical simulations can become extremely useful to correctly address the engine ability to promote adequate flow structures and patterns.CFD analyses of in-cylinder flow motion development and decay are usually carried out using relatively simple yet stable turbulence models, among which k-epsilon and RNG k-epsilon, in conjunction with near-wall algebraic functions, are probably the most popular ones. Nevertheless, it is widely recognized that such models show deficiencies in correctly capturing complex rotating flow structures such as those dominating in-cylinder flows of compression ignition engines. While waiting for the diffusion of more refined approaches such as Large Eddy Simulation (LES) or Detached Eddy Simulation (DES), the use of better performing RANS models such as Reynolds Stress and k-omega SST, in conjunction with adequate resolution of the near wall flow, would be recommendable.The paper reports a numerical activity aiming at evaluating the influence of the turbulence model choice on the prediction of in-cylinder flow patterns in a HSDI Diesel engine for automotive applications. Three different engine operations are investigated, i.e. two part-load / low engine speeds ones and a full-load / peak power engine speed one. For each engine condition, both wall-function k-epsilon and low-Reynolds k-omega SST models are used, and for the last one two different near-wall grid refinements are adopted. In order to better assess the predictive capabilities of the models, analyses are preliminarily compared to experimental measurements for both a simplified engine-like geometry and flow structure available in literature and steady flow-bench operations of the actual engine head.