In order to improve the efficiency of voltage stability assessment （VSA） for large power grids， a three-stage progressive contingency screening and ranking method is proposed. In stage 1， the key contingencies are identified from an initial list of contingencies by defining the importance index of the branches. In stage 2， the critical contingencies are identified from the obtained key contingency set by combining sensitivity-based load margin estimation method and the minimum singular value method. In stage 3， the unstable contingencies are identified and ranked with contingency parameterization continuation power flow. Other critical contingencies are ranked according to their load margins obtained by using curve fitting method. The proposed method is successfully applied to an off-line VSA tool in CSG. The results show that it can be effectively applied in the practical large-scale power systems.