Distributed, self-organized and infrastructure-independent nature of MANET makes it vulnerable to various kinds of attacks. Among them, a threat caused by black hole attack is more prominent. A trust management scheme is a common way to detect and isolate the compromised nodes while a cryptography mechanism shows a failure facing to inner attacks. The previous work proposed two simple models along with trust computation methods to filter out black hole nodes using the benevolent ones and achieved better performance against pure black hole attacks. In this study, a method to prevent gray hole attack, one kind of black hole attack, which is more cunning and clandestine attack is proposed. The neighbour observing model based on watchdog concept is focused on the direct trust value (DTV) to detect single black hole attack. Historical evidence is also taken into consideration against gray hole attacks. Moreover, the neighbour recommendation model that companied with indirect trust value (ITV) is used to detect the cooperative black hole attacks. On the other hand, the proposed method tries to balance the trust values from different neighbour nodes to mitigate the damage by their deceiving. The simulation results demonstrate the advantages of proposed method by punishing malicious actions to prevent the camouflage and deception in the attack.