Abstract: This paper presents an inference system dealing with uncertainties, which is included in EXDASS. Based on expert knowledge estimating and predicting uncertainties/risks, the system provides some supports to decision making through a man-computer dynamic interactive procedure. The paper illustrates the possibilistie evidence theory, which makes up the theoretical foundation of the expert knowledge representation and the inference mechanism. The design priciples and the framework of the system are expounded with an application case, especially, some self-learning features of the system are discussed.