Predicting aircraft dynamics and vibration loads at components' interfaces is a key task for ensuring a robust design and development of the product. Usually, whereas in the case of isolate components the dynamic behavior can be predicted quite accurately, when several components are assembled through discontinuous junctions the predictiveness of a model decreases. The junctions, whose mechanical properties are seldom well characterized experimentally, often introduce nonlinearities in the loads' path. Additionally, their behavior is intrinsically uncertain and as a consequence, the dynamic response of the connected structures becomes stochastic. We propose a sample-based approach which aims to cope with both aspects, nonlinearities and uncertainties, and can be split in two main tasks. First, the computational cost of each deterministic simulation is minimized considering that the global nonlinear behavior depends on localized sources of nonlinearities at the interfaces. Second, the uncertainties are propagated through the model by a non-intrusive method based on Sobol's low discrepancy design. Attention is paid to the global sensitivity indices, which are estimated by creating a meta-model based on Polynomial Chaos Expansion. An industrial application considering an aircraft component whose dynamic behavior is affected by uncertain free-plays at its interfaces is presented.