In dental treatments where metal is indispensable material and dental implants require precise structural measurements of teeth and bones, the ability of CT scanners to perform Metal Artifact Reduction (MAR) is a very important yet unsolved problem. The increasing need for dental implants is raising the demand for a conebeam CT. In this paper, an MAR method of the Metal Erasing Method (MEM) is extended to three dimensions. Assuming that metals are completely opaque to X-ray, MEM reconstructs metals and other materials separately, then combines them afterward. 3D-MEM is not only more efficient but performs better than the repetition of MEM, because it identifies metals more precisely by utilizing the continuity of metals in the third dimension. Another important contribution of the research is the application of advanced binarization techniques for identifying metal-corrupted areas on projection images. Differential histogram techniques are applied to find an adequate threshold value. Whereas MEM needs to identify metals on a sinogram that covers the all rotation angles with a single threshold value, identifying metals on each projection image with an individual value is an important benefit of 3D-MEM. The threshold value varies per projection angle, especially by the influence of the spine and scull, that are objects outside of the field of view. The performance of 3D-MEM is examined using a subject who has as many as 12 pieces of complex metals in his teeth. It is shown that the metals are successfully identified and the grade of metal artifact has been considerably reduced.