Vessel lumen centerline extraction is an important issue for the intra-operative guidance of endovascular instruments; furthermore, vessel centerline is often used as a reference position in many hemodynamic studies, especially in carotid arteries. In this work we propose an innovative method for the extraction of carotid vessels centerline from three-dimensional Color Doppler ultrasound images. The method was tested on carotid Color Doppler images of eighteen healthy subjects and validated by calculating the Euclidean distances between the centerlines detected by the algorithm and those manually annotated by two experts in the corresponding original US volumes. The results show that the proposed approach can accurately estimate the actual centerline with an average error of 1.08 ± 0.54 mm. Furthermore, the method is completely automatic and therefore suitable for the aforementioned purposes.