Within the scope of restoration projects, structural models are used to better understand the structural performance of historical masonry buildings, as well as to design new restoration interventions. The simulation of the structural response in historical buildings, from a mechanical point of view, is a complex problem. It includes complex geometry of structural elements (buttress, pier, wall, arch, vault, flying-buttress, etc.), their connections, mechanical properties resulting from the varying texture of masonry and additionally complicated by anisotropy due to present damage. Ageing and a long history of construction, loads and interventions, may be only partially described or even unknown. Furthermore, the determination of material properties entails several uncertainties and presents practical testing problems. Under these conditions, visual observations of cracks, observation of the masonry texture, measurements of complex geometry, measurement of the axial force in the iron ties or the state of stress in masonry, are important evidence for making an approximation of the real mechanical state. Structural modelling must take these evidence into consideration, here called experimental observation. A bi-directional connection between structural models and experimental observations is needed. Here, the choice is made to gather the most information possible in order to setup a model and to calibrate this in relation to its performance in reproducing the phenomena observed in-situ. The first part of the thesis is focused on an understanding of the three-dimensional geometrical complexity of historical masonry buildings, comprising not only the visible parts but also the hidden internal divisions and connections of the structural elements (typically related to material properties also). An innovative procedure was created to develop a geometrical model with a close relation to the real building structure. The aim being to reach a good approximation of real geometry and a good possibility of creating its discretization into finite element model. The procedure starts with the acquisition of the geometry (photogrammetric measurements and observations), data elaboration and use of parametric procedures to generate complex parts of the system (ie nodal zones, vaults). The data elaboration considers shape identification algorithms implemented in Matlab, while the parametric model is implemented in IronPython programming. The model is finalised for use in a three-dimensional with solid elements, numerical finite element analysis of the structural response. The same geometric model can also be applied for analytical solutions. A discussion on the approximation is developed in two case studies (the Cathedral of Milan and the Church of San Basssiano in Pizzighettone). Overall, the results of this part create a link between the observed geometry, the understood construction technology and the structural models. The developed procedure, reduces the cost of developing complex three-dimensional solid structural models. It consists of a reduction of generation time and better meshing of solid models with hexahedral elements. The second part of the thesis focuses on the development of detailed continuum and discrete numerical models, starting from the previously created geometric models. Two types of models were created in relation to the physical dimensions of the considered domain: a) Small-scale, and b) large scale. The importance of developing complex numerical models on both scales is demonstrated in different case studies, by the type of phenomena and structural response that can be predicted. a) small scale model; the case study of the quadripartite vaults in …