The presence of inherent length scales in composite materials, coupled with the development of defects at the microscale, requires the adoption of a multiscale approach to analyse the mechanical performance of these structures at the application level, namely the macroscale. One of the most prominent approaches for predicting the effective response of the microscale model is the classical homogenisation via Hill-Mandel Principle [1], which couples the scales based on equality of work done at the local macroscale element and a representative volume element (RVE) of the microscale. This work presents a RVE approach in which different scales are coupled by using Bayesian inference [2]. This approach allows two different descriptions of mechanics on representative scales, e.g. discrete on the microscale and continuum on the macroscale, for example. The method is based on identifying constitutive laws of the macroscopic phenomena given information on the microscopic scale. Next to this, we present an approach to account for inherent variation of the microscopic constituents in geometric or material sense, model errors, and loss of resolution due to employment of RVE. This method will be demonstrated for a few numerical examples in 2D and 3D regime.