In this paper, Haar Transform (HT) is used in the fault diagnosis of rotating machinery, and the "Impulse Sharpness" is presented as a diagnostic index. At present, Fourier Spectrumanalysis is most widely used. Compared with FFT, HT is more rapid in computation and more effective in discrete approximation. It's very suitable for the extraction of pulses in the signal. However, HT has some shortcomings. It's greatly affected by the starting point and length of the sample. Here, the authors present a method to improve the stability and comparability of Haar Spectrum. The fault imitating test of rolling bearing is carried out, and the results obtained have verified the sensitivity of Haar Spectrum and Impulse Index to the fault.