The traditional deterministic optimal design is mostly based on meeting regulatory requirements specified in impact standards, without taking the randomness of the impact velocity and angle at the real world situation into consideration. This often leads to the optimization results that converge to the boundary constraints, thus cannot meet the reliability requirements of the product design. Structure members of B-pillar (e.g. inner panel, outer panel, and the reinforcing plate) play a major role in the side impact safety performance. This paper dealt with optimization of B-pillar by considering its dimensions and materials as the design variables, and the impact velocity and angle from real-world traffic accident conditions as the random variable inputs. Using a combination of design of experiment, response surface models, reliability theory and the reliability of design optimization method, a B-pillar was constructed based on the product quality engineering. Comparing the results predicted by traditional deterministic optimization and reliability optimization methods, reliability optimization results met the established crash-worthiness objectives better, thus increasing reliability of B-pillar design considerably.