The co-registration of complex SAR images is a key problem in SAR interferometry, especially when the relative acquisition geometry is not known with sufficient accuracy and the approximate terrain elevation changes cannot be considered a priori. This article presents a new multi-stage co-registration strategy using dynamic programming technique for the globally consistent matching. In the primary stage, coarse co-registration results are obtained based on the cross-correlation algorithm which can always give a robust answer. Then in the following fine registration stages, scaling factor is used to correct the systematic misregistration errors in the range direction while a pixel by pixel dynamic programming procedure is used to give a smooth offset-surface of the azimuth direction, which is usually tied to topography. The dynamic programming technique considers the global compatibility of the matching results and optimizes the entire scan area rather than search for the optimum for each point separately, and thus a more smoothly consistent result. Another advantage of this approach is that it gives estimates for co-registration based on a pixel by pixel basis. Subpixel accuracy is achieved and both spaceborne and airborne interferometric data of DLR are used to show the potential of the proposed co-registration strategy.