Abstract: In fault diagnosis,vibration signals include noise signal and the gear box signal. In order to separate the useful gear signals so that the correct diagnosis of a gear box failure is done,a new method based on the negative entropy algorithm of FastICA was proposed so as to separate gear box vibration signals. The method analyses the vibration signal of domain and spectrum, getting separated signals of their own. The results show that the method can not only help to correctly judge the failure characteristics, and but also enhance the analysis strength of the fault signal,and diagnosize the form of its failure,which indicates that the method can be used as the pretreatment method of the fault diagnosis.