Aimed at the problems that in the traditional gear fault signal processing, the sensitive characteristic information is easily lost in the process of the single signal processing , a formal sampling wavelet (MUDW) fusion pretreatment method of multipoint gear fault vibration signal is proposed to make full use of the characteristic information of each signal contained to reduce the influence of noise and interference elements. First of all, each channel signal is preprocessed by using the time synchronous average (TSA), and then the morphological undecimated wavelet framework signals are decomposed, and each decomposition level approximate signals contribution to the fault is characterized to raise the proportion of useful approximation signals by adopting the correlated kurtosis (CK) and the weighted operation form. On the basis of this, each decomposition level signal is fused by the established fusion rule, improving the reconstructed 〖JP2〗signal characteristic information. Finally, the experiment of gear crack fault shows that the method can restrain noise well, and the meshing frequency and its double frequency are prominent. The fusion effect is ideal.