We consider a mesoscale modeling of a complex microstructure for which the medium is modeled with apparent properties represented by an elasticity-tensor random field. A complete methodology is proposed for the experimental identification of the random field at mesoscale (1) using image field measurements at macro- and mesoscales, (2) introducing two associated boundary value problems for which the change of scale is performed in the context of stochastic homogenization and (3) using statistical inverse methods. A validation will be presented for cortical bone with experimental simulations which measure displacement (or strain) fields at the macro- and mesoscales for a given sample submitted to a given load.