In ray tracking, acceleration structure is indispensable for reducing computation time of ray-geometry intersection. Finding an efficient and high quality acceleration structure is still an active area of research. The top-down approach is widely used in building acceleration structure, such as BVH. However, due to its limitation, the result tree is not always optimal. This paper presents a bottom-up approach to build an effective BVH which is a hybrid between BVH and Space Partitioning. An improved surface area heuristic cost function and multiple parameters have also been used to optimize this model. Experiments show that the approach mentioned above is more effective than the conventional BVH and KD-Trees. It can realize 5%-10% performance improvement. Furthermore, the way to reduce the building time has been investigated in this paper.