The registration of pre- and post-earthquake synthetic aperture radar (SAR) images is a challenging problem.The difficulty lies in that the variform objects to be registered often have complex deformations.To solve this problem,this paper proposes a new registration approach based on rob ust weighted kernel principal component analysis (KPCA).We show how the variform objects of pre- and post-ea rthquake can be precisely registered using their robust kernel principal components (RKPCs).The contribut ion can be divided into three parts.Firstly,a robust weighted KPCA (RWKPCA) method is developed,which can n ot only capture the common RKPCs of the variform objects of pre- and post-earthquake,but also act as the criterion for outlier detection. Secondly,based on the projections on common RKPCs,the similarity measure of th e features of the variform objects of pre- and post-earthquake is defined,and thus the matching results c an be obtained.Finally,a variform objects registration approach is derived from the defined similarity measure and the matching results.Two experiments are conducted on the SAR image registration in Wenchuan earthquake, and the results show that compared with the existing methods,our method is more effective in capturing th e common RKPCs of the variform objects of pre- and post-earthquake,and thus ha s a better registration result.