Stratocumulus clouds have an important impact on the global net radiation budget. The lifecycle of these Stratocumulus clouds is strongly influenced by turbulent mixing, in particular near the cloud top where dry and warm inversion air is mixed into the cloud (entrainment). We investigated the simulation of vertical mixing by a turbulence scheme on basis of a prognostic turbulent kinetic energy (E) and a diagnostic lengthscale (l). The behavior of this E-l scheme is evaluated with observations of Stratocumulus during ASTEX, and a comparison is made with the results of LES models for the same case. It is shown that the E-l model is well capable of reproducing the main features of vertical mixing and entrainment in the LES models. This is the case with a high vertical resolution of 25 meters and a short time step, even with a relatively simple formulation for the turbulent length scale. However, it appears that the simulation degenerates rapidly on coarse temporal and spatial resolution. In particular for an operationally feasable vertical resolution of the order of 200 meters, the model did not behave well. At such resolution, entrainment appears to be predominatly related to the Eulerian (gridbox) representation of the cloud, and not to the physics of the turbulence scheme. This gridbox representation of the cloud prevents the cloud to descend due to the subsidence, and therefore generates an entrainment rate that balances the subsidence rate. An unphysical dependency of the entrainment rate on the subsidence rate results. A simple, and general, conceptual model for this behavior is presented. Finally, the general relevance of the results for large scale atmospheric models is discussed.