Inspired by the interest in agent's preference model, a new preference system called AF preference system is presented in this paper, which is able to describe the behavior of human beings. It combines the rational inference of preference offered by Michael Freund with classical inductive logic and probability logic provided by Adams. That is, relying upon the classical logical system of Adams, the logical reasoning, the general (commonsense) reasoning based on preference inclination of human beings, and the inclined reasoning based on subjective hopes of human beings are combined. It is therefore an organic system. Moreover, in this system the method for constructing the rational preference relations of agent and the process of rational inference are provided. And finally, an example is given to show that the system can fully guess an agent's undisclosed preferences according to the incomplete information on his preferences. Such system provides a mental model for any rational agent. Once the inclination of agent's preference changes, this model can adjust itself accordingly, without the need of doing radical change. It is in some sense a hierarchy of the preference system of Freund and so it is particularly fit for the inductive manner of human mind. It has robustness and utilitarian value.