This paper aims at accounting for the uncertainties because of material structure and surface topology of micro-beams in a stochastic multi-scale model. For micro-resonators made of anisotropic polycrystalline materials, micro-scale uncertainties exist because of the grain size, grain orientation, and the surface profile. First, micro-scale realizations of stochastic volume elements are obtained based on experimental measurements. To account for the surface roughness, the stochastic volume elements are defined as a volume element having the same thickness as the microelectromechanical system (MEMS), with a view to the use of a plate model at the structural scale. The uncertainties are then propagated up to an intermediate scale, the meso-scale, through a second-order homogenization procedure. From the meso-scale plate-resultant material property realizations, a spatially correlated random field of the in-plane, out-of-plane, and cross-resultant material tensors can be characterized. Owing to this characterized random field, realizations of MEMS-scale problems can be defined on a plate finite element model. Samples of the macro-scale quantity of interest can then be computed by relying on a Monte Carlo simulation procedure. As a case study, the resonance frequency of MEMS micro-beams is investigated for different uncertainty cases, such as grain-preferred orientations and surface roughness effects. Figure 1. Homogenization-based multi-scale method: (a) first-order homogenization for classical macroscale continuum and (b) second-order homogenization for macro-scale Kirchhoff-Love plates.fields would require the nested resolution of meso-scale problems during the structural scale analysis, leading to a prohibitive cost. Local effects can also be treated using Monte Carlo simulations:The brittle failure of MEMS made of a poly-silicon material was studied by considering several realizations of a critical zone [24] on which the relevant loading was applied. An alternative to these approaches is to introduce in the stochastic multi-scale method a meso-scale random field, obtained from a stochastic homogenization, which is in turn used as material input by the stochastic finite element method at the structural scale. In order to ensure objectivity, the size of the (structural scale) stochastic finite elements should be small enough with respect to the (spatial) correlation length of the meso-scale random field [8], the latter depending on the size of the SVEs. In order to define a meso-scale random field, statistics and homogenization were coupled to investigate the probability convergence criterion of RVE for masonry [25], to obtain the property variations because of the grain structure of poly-silicon films [26], to extract the stochastic properties of the parameters of a meso-scale porous steel alloy material model [27], to evaluate open foams meso-scale properties [28], to extract probabilistic meso-scale cohesive laws for poly-silicon [29], to extract effective properties of random two-phase composites [30], to stud...