The exact location of intrahepatic vessels in relation to a tumor is an important issue in oncological liver surgery. A sufficient blood supply via the portal veins and blood drainage via the hepatic veins has to be ensured after tumor resection. Computer-aided preoperative surgery planning based on CT or MR data can help to achieve this aim. Intraoperatively the surgeon can be supported by an 3D ultrasound-based navigation system to transfer the plan accurately on the patient in the OR. In this case the preoperative vessel models have to be registered to the intraoperative 3D B-mode or Powerdoppler ultrasound data of the vessels. The problem is, that portal and hepatic veins are imaged in two consecutive acquisitions: In the portal venoues (PV) phase and in the late venous (LV) phase of the contrast enhancement. Accurate registration of PV and LV phase has to be performed in order to compensate for different respiration states of the patient during image acquisition. The aim of this contribution is to investigate the amount of misregistration and how well it can be improved by automatic intensity-based registration methods.