In this study, the trabecular bone was treated as a composite material that consists of a bone matrix weakened by ellipsoidal pores. Under the hypothesis that all information concerning the local properties and the microarchitecture are “encrypted” in the apparent properties of a given volume element (VE) of the bone, a method of retrieving these data was proposed. Software based on a genetic algorithm, combined with the incremental scale transition method was developed to this end. To test the approach, µCT measurements of four bone samples were performed providing their real micro-architecture. Tensors of apparent properties of the samples were next computed by numerical (finite element) homogenization method for a large range of the elastic properties of trabeculae. They were considered as the fitness function for the proposed algorithm. Very good agreement was found between the obtained and target values of the apparent elastic properties of the samples and volume fraction of pores. The approach is fast and accurate enough in comparison to the finite element homogenization. As an auxiliary result it was shown that the anisotropy of apparent elastic properties is mainly related to the microarchitecture of the bone, not to the intrinsic properties of trabeculae.