More than 1 million people die in crashes on the worlds roads and many millions are seriously injured each year. According to the studies: Run-Off-Road accidents (ROR), i.e. the vehicle run-off the road into the roadside and has at least one collision with either roadside equipment or the roadside itself, 'represent about 10% of the total road accidents, while 45% of all fatal accidents are ROR'. Vehicle Restraint Systems (VRS) are the infrastructures installed on the road to provide a level of containment for an errant vehicle. Safety barrier is 'continuous VRS installed alongside, or on the central reserve, of a road to prevent errant vehicles from crashing on roadside obstacles, and to retain them safely'. Statistic results show that 'the existence of protective barriers on road can reduce fatalities up to a factor of 4 when compared to collisions against other road obstacles. The life-saving performances of a VRS depend on the design of the device. Standards such as EN1317 normalized the impact conditions under which a design of VRS must be tested by crash tests, and defined the criteria for performance evaluation of a design. While a VRS cannot really be optimized: Multi-criteria exist for performance evaluation of a VRS and all the criteria cannot be optimized in the same time; the impact conditions of the VRS with the errant vehicle are numerous; uncertain factors of the VRS may degrade the performances of a design. The thesis aims to define an approach that can serve: sensitivity analysis (SA) and robust design of the VRS; Enrichment for the existing standards in the design of VRS. The case of a safety barrier is specified in the study: a safety barrier has been test experimentally, the program Ls-Dyna was used for crash simulation of the device; considering properties of the crash model, efficiencies of different SA methods were studied and influences of the critical factors whose uncertainties contribute the most to the instability of the barrier were quantified with the selected SA approaches; considering the uncertainties of the critical factors, Multi-Objective robust optimization of the tested barrier were realized; under different impact conditions, crash simulations of the optimized barrier were carried out to evaluate its performances in the real crash accidents. The approaches presented in the article can be useful for the design of other VRS or more broadly, other complex engineering systems. Hopefully, the robustness analysis and generalization analysis (i.e. performance evaluation of the VRS under different impact conditions) of the safety barrier could enrich the standards for the design of VRS.