We propose a method based on coherence maximisation based on Thagard's theory of coherence to model rational behaviour. We show that the traditional analysis of behaviour based on utility functions can be emulated using our approach and prove that the maximum element of a preference ordering is the same as that found by coherence maximisation over an automatically generated coherence graph. In addition, it is easy and natural to model the dynamism and uncertainty of the beliefs of a rational agent with this approach. We illustrate the theorem by modelling the prisoner's dilemma.