Abstract: In order to solve the problem that a single method is difficult to deal with large scale,multi-variable data in fault diagnosis,a multi-variable decision tree construction method combining principal component analysis with rough set theory is proposed.Firstly,the method uses principal component analysis to make dimension reduction and remove noises for the historical data and attempts to get the decision-making information that consists of principal component variables.Secondly,the method presents attribute selection and sample set measure for the decision-making information by nuclear properties and relative generalization concept in the rough set theory to construct multi-variable decision tree,on this basis,establishes diagnosis rules repository.Finally,by use of a shafting vibration fault analysis example on steam turbine generator units,the validity is demonstrated.Compared with other methods,this method has the advantages of small scale and is easy to extract diagnosis rules.