An incipient mechanical fault detection method, combining multifractal theory and Mahalanobis-Taguchi system (MTS), which is based on statistical technology, is proposed in this paper. Multifractal features of vibration signals obtained from machine state monitoring are extracted by multifractal spectrum analysis and generalized fractal dimensions. Considering the situation of mass samples of normal mechanical running state and few fault states, the feature parameters corresponding to different mechanical running states are further optimized by a statistical method, based on which incipient faults are subsequently identifi ed and diagnosed accurately. Experimental results proved that the method combining multifractal theory and MTS can be used for incipient fault state recognition effectively during the mechanical running process, and the accuracy of fault state identifi cation is improved. An incipient mechanical fault detection method, combining multifractal theory and Mahalanobis-Taguchi system (MTS), which is based on statistical technology, is proposed in this paper. Multifractal features of vibration signals obtained from machine state monitoring are extracted by multifractal spectrum analysis and generalized fractal dimensions. Considering the situation of mass samples of normal mechanical running state and few fault states, the feature parameters corresponding to different mechanical operating states are further optimized by a statistical method, based on which incipient faults are more identifi ed and diagnosed accurately. results proved that the method combining multifractal theory and MTS can be used for incipient fault state recognition effectively during the mechanical running process, and the accuracy of fault state identifi cation is improved.