Abstract: Due to the complicated structure of an abdominal CT image and the low segmentation precision leaded by the similarity of adjacent organs, a novel 3D liver segmentation algorithm based on probabilistic atlas and Random Walker methods was proposed. Firstly, a liver position probabilistic atlas was established by some liver region images segmented by doctors; then, an improved automated Random Walker algorithm was proposed and a gray probabilistic atlas improved by Random Walker algorithm was constructed; lastly, a three dimensional liver segmentation was conducted based on the proposed algorithm from abdominal CT image. Experimental results showed that the proposed algorithm could efficiently segment liver region and own good robustness. Comparing with the traditional algorithms, the segmentation precision is improved greatly.