In the wireless sensor networks composed of battery-powered sensor nodes, one of the main issues is how to save power consumption on each node. The usual approach to this problem is to activate only necessary nodes (e.g., those nodes which compose a backbone network), and to put other nodes to sleep. One such algorithm using location information is GAF (Geographical Adaptive Fidelity), and GAF is enhanced to HGAF (Hierarchical Geographical Adaptive Fidelity). In this paper, we study the energy-efficient partition of a 3 dimensional sensor field into cells. Further, we give a theoretical upper bound on cell size for this problem.