This paper proposes a joint relay selection and power allocation scheme for the cognitive radio network based on energy harvesting. Under the condition that the primary user performance is guaranteed, this scheme can significantly improve the energy efficiency of system. This paper considers that the transmitters of primary user and secondary user both conduct energy harvesting. In addition, we consider the imperfect channel status information. First，the secondary user is selected as a relay to assist the primary user to transmit data, while the secondary user collects the energy of the radio frequency. Then， we present the relay selection and power allocation tactics to maximize energy efficiency. Since the convex optimization problem is still a fractional programming, we apply fractional deformation and Lagrange method to solve the problem of optimizing the maximum throughput of the primary network. Based on the proposed scheme, an off-line optimization problem is proposed to optimize the throughput of the primary user, and we apply generalized the Bender decomposition to solve this problem. Numerical results show that the proposed scheme has an obvious advantage in energy efficiency.