Cooperative communication offers a way to obtain spatial diversity in a wireless network without increasing hardware demands. The different cooperation protocols proposed in the literature [1] are often studied under the assumption that all channel state information is available at the destination. In a practical scenario, channel estimates need to be derived from the broadcasted signals. In this paper, we study the Amplify-and-Forward protocol and use the expectation-maximization (EM) algorithm to obtain the channel estimates in an iterative way. Our results show that the performance of the system that knows the channels can be approached at the cost of an increased computational complexity. In case a small constellation is used, a low complexity approximation is proposed with a similar performance.