Medical image fusion plays important roles for accurate clinical diagnosis and pharmacology determination. In this article, the fusion on CT/MRI medical images is carried out based on the curvelet transform due to its high sensitivity to two dimensional edges and curves. After analyzing the sub-band coefficients decomposed from the original images, three methods are put forward to manipulate the transformed coarse and fine scales. Results imply that the maximally selecting the coefficient at every scale from the original images is optimal to achieve more effective fusion than the combined methods as maximal selection at fine scales but weighted averaging or simply averaging coefficients at coarse scales. For the latter two methods, the weighted averaging at coarse scale is better. More MRI information extraction is important for the fusion performance, since the MRI image has more details than CT. The algorithms proposed in this article can be integrated into the multi-modal medical imaging instrument to acquire higher accuracy for clinical and pharmacology decision.