Screening similar historical fault-free candidate data would greatly affect the effectiveness of fault detection results based on principal component analysis (PCA).In order to find out the candidate data,this study compares unweighted and weighted similarity factors (SFs),which measure the similarity of the principal component subspace corresponding to the first k main components of two datasets.The fault detection employs the principal component subspace corresponding to the current measured data and the historical fault-free data.From the historical fault-free database,the load parameters are employed to locate the candidate data similar to the current operating data.Fault detection method for air conditioning systems is based on principal component.The results show that the weighted principal component SF can improve the effects of the fault-free detection and the fault detection.Compared with the unweighted SF,the average fault-free detection rate of the weighted SF is 17.33％ higher than that of the unweighted,and the average fault detection rate is 7.51％higher than unweighted.