The ATMS is a truth maintenance system which determines what nodes are believed in the current context, finds contradictions contained in knowledge base automatically, and retracts them efficiently. Although it provides a good way to manipulate knowledge without uncertainty, it is difficult to deal with knowledge with uncertainty used in expert system. This paper firstly describes a system called EATMS which is an extended version of ATMS to maintenance truth of knowledge with uncertainty, secondly discusses what are contradictions about knowledge with uncertainty, and introduces two concepts called possible nogood and possible good to make hypotheses about current situation in which the given information is not enough for reasoning, and at last gives an algorithm that selects assumptions from possible hypothesist, and updates them when contradictions are found, or the result of reasoning is not safficient one.