The intrinsic low tensile strength of masonry makes existing structures vulnerable to seismic forces, and, for this reason, reinforcing interventions are often required [1]. Here, the attention is mainly addressed to an alternative retrofitting technique, namely the Steel Fiber Reinforced Mortar (SFRM) coating [2]. It involves the application of thin mortar coatings, reinforced with randomly distributed short fibers in the mortar matrix, either on one or both sides of masonry walls. Among the few modelling attempts of this reinforcement method, [3] suggests a homogenization approach, which involves identifying the in-plane equivalent elastic constants of a Cauchy continuum. On the other hand, homogeneous Cauchy models are not suitable for describing the out-of-plane behaviour of these kinds of systems. Accordingly, FE (Finite Element) models have to be employed, incurring a significant computational burden due to the requirement for a highly dense mesh with dimensions at least equal to the thickness of the thin coating. In the present work, an alternative approach is proposed, namely: (i) the unreinforced masonry structures are represented by 3D (both refined and homogenized) FE models; (ii) the reinforcement layers are incorporated through a ‘surface stress’ model, also referred to as the Gurtin–Murdoch model of surface elasticity [4]. Accordingly, the presence of thin layers is modelled by a non-classical boundary condition which gives the surface traction on the substrate in terms of surface stress and inertia. Such a model, usually applied in Micro- and Nanomechanics, is here employed for the first time to the field of civil engineering. In particular, numerical FE simulations in static and dynamic regimes are carried out on various sample systems of coated masonry, such as facades with openings, vaults, walls with variable cross-section. Comparisons with refined finite element models are performed to evaluate the effectiveness of the proposed approach and the accuracy of results, particularly when the reinforcement thickness is small, along with a significant reduction in computational burden.