The appropriate choiche of turbulence closure plays a fundamental role in improving the accuracy of Boundary Layer flow modelling. This is particularly true when dealing with complex terrain, as the current available closures are formulated for flat terrain. Consequently, accounting for the t.k.e. transport and diffusion is needed. Several turbulence models, characterized by different orders of closure, can be found in literature, but a compromise between the complexity of the closure formulation and its appli- cability in mesoscale circulation models is necessary. In this work our interest was focused on evaluating the performances of four turbulence closure models against the EPA-RUSVAL wind tunnel experiment data-set, where a neutral flow over a schematic two- dimensional valley has been reproduced. The simulations of the mean flow and turbulence fields were performed using the fluid- dynamic model CSU-RAMS in its non-hydrostatic version. Two clo- sure models out of them are adopted in RAMS, a Smagorinsky first-order deformation type and the Mellor-Yamada level 2.5, while the third and the fourth ones, a standard E-l model and an E-e-l model respectively, have been introduced by us as alterna- tive options in the code. Comparison is made in terms of vertical profiles of the horizontal and vertical velocity components. Vertical profiles of turbulent kinetic energy were also con- sidered, when the models take into account its prognostic equa- tion. In all the cases we adapted the closure scheme to the characteristics of the flow and turbulence in the wind tunnel. For all the closures, we estimated the constants defining the diffusion coefficients according to the similarity theory and to the surface layer concept for the neutral stratification. The turbulence measured in the wind tunnel and the physical parame- ters of the experiment were also accounted for when possible. Different formulations for the mixing length have been adopted. The results of the simulations were improved by using the values we prescribed. Despite its simplicity, Smagorinsky model gave reasonably good results. The results obtained by the Mellor- Yamada model were not satisfactory: weak or negative velocity values inside the valley and t.k.e. underestimation were found. E-l standard model, parameterising the dissipation rate and the mixing length analogously to the Mellor-Yamada closure, showed a good agreement with observed flow and turbulence profiles. E-e-l model yielded satisfactory results, even if it showed some insta- bilities in the balance between the t.k.e. and its dissipation term.