A decision-fusion scheme is proposed for robust blind channel estimation on the source-relay-destination link in amplify-and-forward cooperative communication networks with selfish relays. The proposed scheme exploits the inherent spatial diversity of such networks to improve the decoding accuracy of the received data, thereby enhancing the channel estimation performance. In addition, a detector is used to identify the existence of selfish behavior at the relay such that the reliability of the received data can be assessed. Having removed the unreliable data where necessary, a least square estimator is applied to determine the channel state information of the source-relay-destination cascaded link. The simulation results confirm that the proposed scheme yields a robust estimation performance in cooperative communication networks with selfish relay nodes.