In modern global networks, principals usually have incomplete information about each other. Therefore trust and reputation frameworks have been recently adopted to maximise the security level by basing decision making on estimated trust values for network peers. Existing models for trust and reputation have ignored dynamic behaviours, or introduced ad hoc solutions. In this paper, we introduce the HMM-based reputation model for network principals, where the dynamic behaviour of each one is represented by a hidden Markov model (HMM). We describe the elements of this novel reputation model. In particular we detail the representation of reputation reports. We also describe a mixing scheme that efficiently approximates the behaviour of a trustee given multiple reports about it from different sources.