Bayesian network is one of the most effective tools for knowledge representation and inference under uncertainties, the structure learning of which is currently a hot topic for research. This thesis proposes a chain-model based particle swarm optimization algorithm for the learning of Bayesian network structure. This algorithm first defines and uses a rule chain model, which contains information about the causality among the Bayesian network nodes, to enhance the quality of the topological sequences searched. It further adds a dynamic weight coefficient to the optimization algorithm for the setting of position of particles to balance the global search with the local search, and to improve the searching capacity of the algorithm at large. Experiments conducted on the general ALARM data set has shown that the new algorithm proposed herein produces better solutions with enhanced rate of convergence compared with the existing Bayesian Network Structure Learning Algorithm Based on Conditional Independence Test and the Ant Colony Optimization (I-ACO-B).