An investigation of turbulence models prediction capacity based on classical statistical modeling and large eddy simulation (LES) is presented. The classical statistical modeling (average of Reynolds) was analyzed, by investigating the solution of complex flows, as, for example, the turbulent flow past a backwardfacing- step (backstep). Special attention was given to low Reynolds number k-e models and models derived by renormalization group theory (RNG). The behavior of the different terms in the turbulent kinetic energy equation in the near wall region was examined in details, specially the pressure diffusion term. It was evaluated the importance of the correct modeling of the pressure diffusion term on the predictions of the low Reynolds number models, in recirculating flows. A few models, proposed in the literature for the pressure diffusion term, were also evaluated theoretically and numerically. The prediction capacity of large eddy simulation (LES) technique was also investigated. The ability of Smagorinsky model to predict complex limited wall flows was analyzed from a computational standpoint. The finite-volume method was employed to integrate both the Reynolds average and LES equations. The fully developed turbulent channel flow was solved in two- dimensional and three-dimensional numerical simulations. The turbulent flow over a backward-facing-step was computed exclusively in a twodimensional manner, while the fully developed turbulent flow in a straight square duct was simulated in a three-dimensional manner. The results were compared with those obtained by the low Reynolds models, analyzing the cost-benefit relation.