A new method to remove metal artifacts utilizing virtual dual-energy CT image sets generated from monoenergetic CT images and dual-energy CT subtraction is presented in this work. CT images were derived from Optima CT580 (General Electric Company, Fairfield, Connecticut, USA). Optimized conversion model from CT numbers to linear attenuation coefficients (LAC) was applied to calculate an accurate LAC map at specific energy. According to mass attenuation coefficients (MAC) of base materials from the National Institute of Standards and Technology (NIST), a LAC map at another higher energy was obtained, and then a set of CT images was derived from the LAC map, which is at different but a known energy. Then, dual-energy subtraction was applied to remove metal artifacts. Results: Between the CT image sets of virtual high energy and the original, there is no significant difference in STD (standard deviation) (no more than 1.91%), while Merror (a parameter for quantification of the CT value differences between two images at the same position) varies from 58.83 to 101.6442. CNRs (Contrast-noise-ratio) in dual-energy subtracted CT images are 1.9% higher than those in the CT images processed by polar coordinate transformation. Conclusions: The Dual-energy subtraction is proved to be a better method for reduction of metal artifacts than the polar coordinate transformation scheme. Moreover, the dual-energy subtraction method is based on the reconstructed CT images obtained with a single energy CT scanner, which is more convenient for users not having access to the projection data.