This paper studies an energy-efficient (EE) design for wireless powered mobile edge computing (MEC) system. We introduce a new EE performance metric called computation energy efficiency (CEE), which is defined as the total computation bits (achieved by both local computing and offloading) per energy consumption. Specifically, we study a CEE maximization problem for the wireless powered MEC system, which is non-convex and thus cannot be solved by standard convex optimization techniques. To optimally solve the problem, we divide it into its inner problem and outer problem. Although the inner problem is non-convex, we derive its optimal solution in a closed-form. Then, based on the solution of the inner problem, we optimally solve the outer problem by using the Dinkelbach method. Simulation results show the effectiveness of our proposed scheme in terms of CEE performance and convergence speed.