Segmentation of 3D cerebral vasculature is important for clinical diagnosis. However, many relevant thin vessels are not visible in 1.5T and 3T MRA. With the recent introduction of 7T MRA, images of higher resolution can be acquired, which contain much more thin vessels. We propose a fully automatic hybrid approach for segmenting vessels from 7T MRA images of the human cerebrovascular system. First, thick vessels and most parts of thin vessels are segmented using a 3D model-based approach and, second, missing parts in regions with low image contrast are segmented using a generative Markov random field approach. The performance of the approach has been evaluated using real 3D 7T MRA images.