A fundamental problem investigated in this paper is computational homogenization of two-component particulate and fibrous composites with material characteristics treated as symmetric but non-Gaussian random processes. Homogenization method is based on equity of deformation energies for the real multi-component and the homogenized Representative Volume Elements (RVE) of the given composite structure. The iterative generalized stochastic perturbation technique is used to provide the additional Taylor expansions about expected values and to finally derive analytical formulas for the basic probabilistic characteristics of the effective tensor components. These expansions use polynomial response functions relating this tensor components with real material characteristics determined through the series of the Finite Element Method homogenization tests and the Least Squares Method. Applicability of this stochastic strategy and its Stochastic Finite Element Method implementation are verified for two well-known probability distributions, i.e. the uniform and the triangular one, to model uncertainty in Young modulus of some composite components. Validation of this approach is carried out using two alternative methods – statistical one known as the Monte-Carlo simulation as well as the semi-analytical one using the same polynomial responses as the perturbation approach, but finally processed using classical analytical probability integrals. This study generally confirms applicability of the proposed probabilistic method to homogenize composite structures with linear elastic and isotropic components exhibiting uniform and triangular uncertainties in their elastic properties.