Complex built-up structures such as vehicles have a variety of joint types, such as spotwelds, bolted joints, rubber joints, etc. Rubber joints highly contribute to the nonlinear level of the structure and are a major source of uncertainties and variability. In the general framework of developing engineering tools for virtual prototyping and product refinement, the modeling of the NVH behavior of rubber joints involve the computational burden of including a detailed nonlinear model of the joint and the uncertainties and variability typical of that joint in a full-scale system model. However, in an engineering design phase the knowledge on the joint rubber material properties is typically poor, and the working conditions a rubber joint will experience are generally not known in detail. This lack of knowledge often do not justify the computational burden and the modeling effort of including detailed nonlinear models of the joint in a full-scale system model. Driven by these issues a non-deterministic numerical methodology based on a three-level modeling approach is being developed. The methodology aims at evaluating directly in the frequency domain the sensitivity of the NVH behavior of a full-scale system model to the rubber joint material properties when nonlinear visco-elastic rubber material behavior is considered. Rather than including directly in the model a representation of the rubber nonlinear visco-elastic behavior, the methodology proposes to model the material nonlinear visco-elastic behavior by using a linear visco-elastic material model defined in an interval sense, from which the scatter on the full-scale system NVH response is evaluated. Furthermore the development of a multilevel solution scheme allows to reduce the computational burden introduced by the non-deterministic approach by allowing the definition of an equivalent linear interval parametric rubber joint model, ready to be assembled in a full-scale system model at a reasonable computational cost. By using a commercial finite element code the developed methodology is illustrated through a numerical case-study: the low-frequency dynamic analysis of automotive door weather-strip seals.