Abstract: Rough sets and decision tree theory are introduced in complicated intelligent fault diagnosis system (CIFDS). A rough-decision fault diagnosis model is thus developed to ensure diagnosis precision and speed up the implementation of CIFDS. The model can extract rules directly from reduced decision table. Rough sets theory as a new mathematical tool is used to deal with inexact and uncertain knowledge for pattern recognition. The target is mainly to remove redundant information and seek for reduced decision tables. As a quickly learning theory and classification tool, decision tree is used to extract rules directly from reduced decision table so as to acquire satisfactory result. An example is given to show how to apply the intelligent fault diagnosis to RH-KTB vacuum metallurgical system. The effectiveness of the algorithm is therefore proved through the exemplification.