Numerical simulation is now a more and more commonly used tool to investigate flows inside IC engine combustion chambers. A key parameter of these codes is to simulate accurately turbulence since it describes the flow structure and turbulence results are inputs for combustion and spray models. In usual industrial codes, in-cylinder flows are calculated with the standard k - ε model. Nevertheless, deficiencies of this model for engine flow simulations are now well known. The turbulence anisotropy due. to the volume variation in the cylinder axis direction can not be taken into account by the k - ε model since it is based on the hypothesis of a single turbulent velocity scale. Moreover, this model is deficient to simulate swirling flows and it poorly predicts recirculating zones (these two kinds of flows are frequently encountered in combustion chambers).In order to improve the turbulence modeling in IC engines and to go beyond the eddy viscosity approach, we have adopted a second-moment closure model and implemented the model of Launder et al. , denoted the LRR model, in the Kiva-II code. After a brief description of the LRR model and its numerical implementation, a backward facing step flow calculation is presented. This validation test case points out the advantages of a Reynolds Stress model over an eddy viscosity model.Finally, a simulation of an intake-compression stroke on a single cylinder engine using both the standard k - ε model and the LRR model is described. Two configurations are analyzed, the “standard” case and the case, denoted the “swirl” case, with a deflector on the intake valve to generate a swirling flow. Numerical results are compared with one and two point LDA measurements. The velocity fields predicted by the two turbulence models are quite different, particularly at the end of the compression stroke. Near TDC, only the LRR model is able to capture the turbulence anisotropy in the cylinder and more precisely in the spark ignition zone.