Abstract. We present a multi-scale approach for non-rigid image registration of tubular tree-like structures such as vessels. Therefore, we consider a Gaussian and a combined morphological and Gaussian scalespace. Similar to a multi-level framework, we solve a sequence of registration problems on several scale-space levels using a standard variational approach for non-liner volumetric image registration. The overall idea is to avoid locally ambiguous mappings between parts of the images by removing morphological details but also finding a global optimal solution by spreading remaining local information using Gaussian scaling. We successfully tested our method on registration of 3D CT and ultrasound images of hepatic vessels.