The main characterisrics of ad hoc networks are the lack of predefined infrastructure and the dynamic topology. These characteristics present some new security vulnerabilities for this emerging networking paradigm. Usually, security in ad hoc networks is handled through authentication and encryption. This can be considered as a first line of defense, however, this remain inefficient against some other kind of attacks such as malicious packet dropping. The purpose of this work is to provide a mechanism for detecting malicious incorrect packet forwarding attacks. To this end, a trust model extending routing protocols and based on the reputation concept is developed. Our model provides two main functionalities: monitoring the behavior of the neighboring nodes in the network and computing their reputations based on the information provided by the monitoring. This paper also discusses how the reputation information is gathered, stored and exchanged between the nodes, and computed according to the different scenarios. Our mechanism is also validated with some simulation work showing its feasibility, performance and benefits.