We are interested in the construction of stochastic reduced-order models in low-frequency range for dynamical structures which are made up of a stiff master structure and several flexible substructures. In the low-frequency range, this type of structure is characterized by the fact that it exhibits not only the classical global elastic modes but also numerous local elastic modes which cannot easily be separated from the global elastic modes. To solve this difficult problem, a new approach is proposed for constructing a reduced-order model adapted to this case. This method consists in introducing two unusual eigenvalue problems which provide a basis of the admissible displacement space written as the direct sum of a global displacements space and of a local displacements space. A stochastic reduced-order model is then introduced in order to take into account uncertainties. Two industrial applications are presented. The first one is relative to an automobile and the second one is relative to a fuel assembly.