The occupant and seat belt restraint system were modelled mathematically using multi-body dynamics. The interactions between belt and dummy such as belt slip and energy dissipation were considered in three dimensional mathematical models of 50th percentile Hybrid111 dummy and 3-point seat belt system. In this study, the commercial software DADS (Dynamic Analysis and Design System) was used for dynamic simulation of occupant impact response in sled tests and barrier crash tests. Results from the sled test simulations were compared with those available in the literature in order to validate the mathematical model for occupant behaviour analysis. In order to simulate the occupant behaviour in barrier crash tests, simple models of car interior system and energy-absorbing steering system were also established. And the effect of seat belt systems on occupant protection was evaluated through the occupant behaviour simulation in barrier crash tests. Several factors, which were thought to be closely related to the performance of seat belt systems for occupant protection, were considered. Orthogonal array table was used for a systematic procedure to lay out the factors and conditions in the barrier crash test simulation. The various injury values were investigated in response to the change of seat belt variables such as shoulder belt slackness, lap belt slackness, belt stiffness, retractor performance, and outer D-ring position. The contribution of seat belt variables to occupant protection was analysed by ANOVA (analysis of variance). From this analysis, it was found that seat belt slackness had the most dominant effect on occupant injuries, and seat belt pretension can increase the occupant safety. Through the systematic applications of orthogonal array table and ANOVA to seat belt design, the modified combination of seat belt variables was obtained, and the initial occupant injury values could be decreased.