This paper describes a methodology for probabilistic design modeling and the integration of structural shape and sizing optimization with crash simulation and reliability analysis to improve crash performance while reducing mass of automotive structures. As direct integration of crash simulation, reliability analysis, and numerical optimization can be very computationally intensive, metamodeling methods are used to establish explicit relationships between the responses of interest and variables governing the design. Whereas highly nonlinear responses such as peak acceleration and intrusion distance are modeled using multiquadric radial basis functions, a second-order response surface model based on forward stepwise regression is used to relate structural mass to variations in shape and sizing variables. The design methodology is applied to reliability-based optimization of front side-rail component of a passenger car under full frontal and offset frontal crash scenarios. Changes in cross-sectional geometry of the side rail are controlled through geometric perturbation vectors and associated shape design variables. With responses of the baseline vehicle model known, a series of deterministic and reliability-based optimization problems based on single- and multi-objective formulations are solved with the optimal design responses validated using full-vehicle crash simulations.