Aiming at the limitation of generating cognitive model from fixed cognitive structure, this paper proposes that in the learning process of fixed cognitive elements and not fixed cognitive structure using bayesian inference method and PFNET theory to generate “the optimal Ki set of nodes” and “the optimal Ki cognitive link” according to the learner requirement from the sample information of previous learners, and get the B-P cognitive model; explaining the generation principle of B-P cognitive model and verifying the effectiveness and feasibility of the model by example.