A computational study of flows in 2D model of four-stroke homogenous charge compression ignition (HCCI) engine cylinder was conducted by standard k-? and renormalization group (RNG) k-? models in FLUENTŽ release 6.2. The primary objective was to evaluate the aforementioned turbulence models for simulation of in-cylinder flows. The engine model was a 2D geometry of a cross section of piston-cylinder assembly, passing through the centres of intake and exhaust valves and parallel to the cylinder axis. Simulation was performed for five consecutive engine cycles under cold-flow assumption and with air as the working fluid. It is generally believed that standard k-? model over-predicts the turbulence kinetic energy, k. The results obtained from the fifth engine cycle at bottom dead centre (BDC) of the intake stroke revealed that below the intake valve and in the near-wall region, standard model estimates up to 26% higher values of k. Overestimation of k by the standard model, results in prediction of higher and less accurate values of turbulent viscosity, ? by this model. Subsequently, in-cylinder flows exhibit large rates of flow deformation, and by definition, RNG model predicts more accurate values of ? in rapidly strained flows. Below the intake valve, ? is predicted up to 53% lower by the RNG model. Therefore, RNG model, in comparison to standard model, is a better approach for computational investigation of in-cylinder flows.