Aiming to separate the vibration signals of the gearbox from the other disturbing components, a new multi-channel data fusion procedure, combining the wavelet package independent component analysis (WPICA) and multivariate empirical mode decomposition (MEMD), is proposed for the identification of gear pitting. First, the gearbox vibration source is extracted by applying the WPICA, which is a kind of sub-band decomposition independent component analysis (SDICA) to the multi-channel pump set signals. Second, multi-dimensional IMFs are obtained through the decomposition of MEMD. The mutual information between different IMF matrices in order to implement the formulation of fault sensitive intrinsic mode function (IMF) is calculated at the last step. In case the fault sensitive IMF is found out, the frequency dominated in this IMF is determined as the characteristic frequency of gear pitting. Data analysis shows the efficiency of the proposed procedure.