Abstract-To improve the computer-assisted diagnosis and decision in dentistry, we tested a new method combining Hu moment invariant (HMI) method and extreme learning machine (ELM) to implement the teeth classification in cross-section image of Cone Beam Computed Tomography (CBCT). 160 images were analyzed and 4 categories were recognized. The results showed the sensitivities of incisors, canine, premolar, and molars were 78.25± 6.02%, 78.00± 5.99%, 79.25± 7.91%, and 78.75± 5.17%, better than ANN statistical-significantly.