Based on the comparative analysis of the current DCS software evaluation methods, the authors adopted the BTP7-14 of NUREG-0800 as the reference standard for evaluating AP1000 nuclear power plants. An overall framework based on Bayesian Belief Network(BBN) for reactor protection system software reliability evaluation was therefore presented. The dependency between evaluation model criteria was decided. A quantitative evaluation was performed based on the software evaluation model. Furthermore, a sensitivity analysis method is proposed for identifying the key criteria and the refinement degree of each criterion which will be helpful to make clear the direction of the software quality improvement.