Metal artefact reduction has become a challenging issue for practical X-ray CT applications since metal artefacts severely cause contrast degradation and the misinterpretation of information about the property and structure of a scanned object. In this paper, we propose a methodology to reduce metal artefacts in a three-dimensional industrial cone beam CT system. We develop a registration technique managing the three dimensional data in order to find accurate segmentation regions needed for the proposed algorithm. Through various simulations and experiments, we verify that the proposed algorithm reduces metal artefacts successfully.