The design optimization of valve-spring is achieved under the condition of expected reliability. The restrain conditions about the reliability of static strength and fatigue strength are considered, and then a dual-objective optimal problem of valve spring is modeled to obtain the lightest mass of valve-spring and the minimum error of spring stiffness. The feasibility of reliability condition is discussed based on evidence theory in order to improve the computational efficiency. By means of the combination of evidence, the upper and lower bounds of reliability are obtained, and a substituting model of restrain condition about reliability is proposed based on the obtained bounds of reliability. After the weighted combination of two objective functions is made, the optimization model is solved by using genetic algorithm. An example is given and it shows that the proposed method is effective.