The extraction of effective recognition feature is important for high dimensional date for pattern recognition. In order to solve the problem of low accuracy and high feature dimension of fault feature for the present fault diagnosis methods. A new fault diagnosis algorithm based on exponential regularized null space linear discriminant analysis is proposed on this paper. Null space linear discriminant analysis (NSLDA) shows desirable performance, in this paper, the within class sample scatter is regularized at first in order to avoid the small sample size problem, then we utilize the matrix exponential to the discriminant criterion to enhance the performance. The proposed method integrated the advantages of NSLDA in dimension reduction and RLDA in pattern recognition and effectively improved the accuracy of fault diagnosis. The validity of the proposed method is verified by the instance of the fault diagnosis of a gearbox and engine misfire.