In mobile ad hoc networks (MANETs), the nodes need to cooperate each other to establish multi-hop routes for out-of-range wireless communication. However, some of them may not always behave normally, either behaving selfishly for saving computational resource or maliciously for compromising communication protocols. Regardless of intents, such misbehavior would lead to the degradation of network performance. It is therefore important to design appropriate mechanisms to ensure that network performance could be maintained at an acceptable level in the presence of misbehaving nodes. But the open nature of MANETs makes such designs challenging. Reputation system has been widely recognized as an effective approach, which associates the behavior of nodes with its reputation, which is calculated by specifying and quantifying the observations of interest with respect to predefined performance metrics. More interestingly, the observations can be obtained and integrated from multiple layers, facilitating cross-layer analysis. This paper intends to take a deep look into several well-studied reputation systems and examine their operational characteristics in terms of modeling approaches and redemption techniques, with an objective to identify their capabilities in terms of misbehavior detection coverage and blind spots. Furthermore, such an anatomy allows us to better understand the failure curses of the deployment and operation of reputation systems in MANETs, so as to improve their performance by adopting effective countermeasures.Copyright © 2014 John Wiley & Sons, Ltd.