Restricted by model complexity and high number of model runs for sampling-based Sensitivity Analysis (SA), qualitative SA methods are widely used for parameter study of complex engineering models (i.e. numerical model of high calculation cost with multiple variables) and quantitative SA of such models has always been a challenge. This work explores the approaches for SA of complex engineering systems and carries out quantitative SA of Vehicle Restraint System (VRS): Influential factors of VRS model were identified through screening analysis (qualitative SA with 2-level screening method and Morris method); The non-influential parameters can be fixed and reduction of inputs variables makes it possible for quantitative SA of VRS, Variance-based SA method (SOBOL' indices) was used to quantify the influence of the influential factors, and metamodel technology --- Kriging interpolation --- was used to generate surrogate model to facilitate the quantitative SA. The factors which contribute most for the robustness of VRS were identified and their influences for model uncertainties were quantified by combining screening analysis and Variance-based SA.