Power allocation (PA) and multiple relay selection (MRS) for cognitive relay networks (CRNs) are considered in this paper. In CRNs, the power of each relay node is limited by the interference constraint on primary users. In order to satisfy the interference constraint, we propose a simple PA scheme in which the power of each relay is inversely proportional to the interference channel from the secondary relay to the primary user. That is, the interference constraint is satisfied by allocating more power to the relays which have less interference to the primary users. After satisfying the interference constraint by the proposed PA scheme, we consider the MRS in order to maximize the secondary system capacity. Since the optimal MRS requires the exhaustive search for all possible relay combinations, we propose a low complexity suboptimal MRS scheme based on the relay ordering. By applying the proposed PA and selecting the relays in order of achieving the larger SNR, we can find a relay combination that performs close to the optimal MRS. Some simulation results show that the proposed MRS scheme achieves about 50% throughput gain over the previously proposed schemes when the maximum transmission power of each relay is 20dBW. Also, the proposed suboptimal MRS scheme with much lower complexity performs close to the optimal MRS using the exhaustive search.