This paper deals with a reduction method of models composed of a linear behaviour subsystem which has a high number of eigenmodes in the range of analysis and of a nonlinear behaviour subsystem. Each subsystem has model uncertainties and data uncertainties. Those uncertainties are taken into account using the usual parametric probabilistic approach and the non parametric probabilistic approach. We present a numerical example constituted of a simple system owning all the properties of the systems we are interested in and which validates the proposed methodology.