Ultra-dense networks (UDNs) are expected to be applied for the fifth generation wireless system (5G) to meet the requirements of very high throughput density and connections of a massive number of users. Considering the large amount of small base stations (SBSs), how to choose proper backhaul links is an important problem under investigation. In this paper, we propose a wireless backhaul algorithm to find an effective backhaul method for densely-deployed SBSs and to maximize energy efficiency of the sys-tem. We put forward adaptive backhaul methods of indirect and direct modes. The SBS can select the direct backhaul which con-nects to the macro base station (MBS) directly, or the indirect backhaul which selects an idle SBS as a relay based on the back-haul channel condition. The algorithm also allocates network resources, including the power of SBSs and system bandwidth, to solve the serious interference problem in UDN. Finally, the simulation results show that the proposed wireless backhaul algorithm has desired performance to achieve higher energy efficiency with required data rate.