The aim of this talk is to derive both a model and a numerical scheme for the approximation of multiphase models of Baer & Nunziato types[2, 7]. Such models are averaged models, able to both model interface flows and well mixed flows. They can be obtained by averaging Euler models following ideas developed in [3]. A new averaging method was developed in [5], based on an explicit stochastic model. We will show that the multiscale model obtained contains several known models in some limits (e.g. nonconservative and relaxation terms of [7, 5]), and also that it ensures in general all phasic entropy inequalities. Then, we will also show that the same method can be applied at the discrete level for deriving a numerical scheme, based on the ideas of [1, 4]. This numerical scheme will be proven to ensure also positivity and all the entropy inequalities under CFL conditions that can be explicitly derived. As the model, the numerical scheme is multiscale in the sense that it depends on a parameter modeling the local topology of the flow. Last, a simple model for the topological parameter will be discretized, and numerical results with this micro-macro model will be presented.