This paper presents an automatic registration scheme to register geometric distortion in satellite images. A novel feature based matching scheme is proposed which establishes correspondence between the corner points in the reference and target images either by correlating the intensity values around a circular neighborhood of these corner points or by exploiting the relative orientation of lines connecting these corner points. Affine transformation model is used to estimate transformation parameters. Resampling is carried out by nearest neighborhood interpolation. The registration process is automatic and can efficiently serve as preprocessing stage for multitemporal analysis, image fusion, and image mosaicking and change detection. The effectiveness of the algorithm has been verified by an intensive experiment on a large number of real images. Experimental results reveal high supremacy of the proposed registration method.