Cone-beam breast computed tomography (CB Breast CT) can easily detect micro-calcifications and distinguish fat and glandular tissues from normal breast tissue. However, it may be a challenging task for CB Breast CT to distinguish benign from malignant tumors because of the subtle difference in x-ray attenuation. Due to the use of polyenergetic x-ray source, the x-ray and tissue interaction exhibits energy-dependent attenuation behavior, a phenomenon that, to date, has not been used for breast tissue characterization. We will exploit this spectral nature by equipping our CB Breast CT with dual-spectral imaging. The dual-spectral cone-beam scanning produces two spectral image datasets, from which we propose a nonlinear dual-spectral image fusion scheme to combine them into a single dataset, thereby incorporating the spectral information. In implementation, we will perform dual-spectral image fusion through a bi-variable polynomial that can be established by applying dual-spectral imaging to a reference material (with eight different thicknesses). From the fused dataset, we can reconstruct a volume, called a reference-equivalent volume or a fusion volume. By selecting the benign tissue as a reference material, we obtain a benign-equivalent volume. Likewise, we obtain a malignant-equivalent volume as well. In the pursuit of the discrimination of benign versus malignant tissues in a breast image, we perform intra-image as well as inter-image processing. The intra-image processing is an intensity transformation imposed only to a tomographic breast image itself, while the inter-image processing is exerted on two tomographic images extracted from two volumes. The nonlinear fusion scheme possesses these properties: 1) no noise magnification; 2) no feature dimensionality problem, and 3) drastic enhancement among specific features offered by nonlinear mapping. Its disadvantage lies in the possible misinterpretation resulting from nonlinear mapping.