Traditional database security mechanisms focus on either protection or prevention. However, in practice all attacks are not avoidable. To solve this problem, Intrusion Tolerant Database Systems (ITDBs) were introduced. An ITDB uses new generation database security mechanisms to guarantee specified levels of data availability, integrity and confidentiality in the presence of successful attacks. A key part of an ITDB is the intrusion detection (ID) which informs the system about attacks. One of the problems in using ID is the false alarm that will lead to the reduction of the "availability" or "integrity". This paper presents an intelligent method to control false alarm. In this method, we will use the significance degrees of data objects to determine the anomaly threshold adaptively, as the "availability" and "integrity" required by the data objects are satisfied.