Aiming at the uncertainty of the evaluation information entered in the requirement satisfactory degree evaluation of capability of ground air-defense systems, a method based on directed evidential network with conditional belief functions is proposed, which uses expert knowledge and historical information to carry on evidential network parameter modeling, i.e., the conditional belief functions are used to describe the relationship between the degree of the lower level capability satisfying the demand and the degree of the upper layer capability satisfying the demand. On this basis, the evidential network is used to predict the belief distribution of capability of the system satisfying the demand. The feasibility and effectiveness of the method are analyzed by an example.