The aim of this study is to develop a new 3D virtual sculpting system by using a hash-based octree data structure with an optimized approach to reduce the huge memory and computation cost and also the tree traversal time. The system first reads the selected 3D virtual object model, which must be in a triangular watertight polygon mesh format. Then both the surface and the interior volume of the virtual object is voxelized in order to generate its volumetric dataset. Afterwards, a hash-based data structure with a novel optimized approach is applied to the dataset. After each sculpting process applied to the virtual object, its surface is locally reconstructed and a feedback force is calculated synchronously. Finally, to give the user a realistic sense of interaction, the feedback force is sent back via a haptic device. Creation durations of a traditional octree, a hash-based octree, and our optimized hash-based octree are compared. Tree traversal and surface reconstruction times of our optimized hash-based octree approach are also compared with the pointer-based one in off-line and real-time while sculpting a 3D model. It is shown that the memory requirement and durations of our optimized approach are the lowest.