In this work, we consider channel estimation for an amplify-and-forward (AF) two way relay network (TWRN), where two terminal nodes exchange information via a single relay node in between. A new concept of channel estimation by performing the power allocation at the relay node is introduced. As an example, we consider the maximum likelihood (ML) channel estimation at relay node and derive the power allocation factor such that the average effective signal-to-noise ratio (AESNR) at the terminal nodes is maximized. Nonetheless, the idea of using power allocation at relay node can be straightforwardly extended to more general scenarios. The simulation results show the advantages of the proposed method over the existing techniques.