Aiming at the difficulty to accurately extract the gearbox fault features under strong background noise in oilfield, an intelligent diagnosis method of gearbox faults based on deep belief network (DBN) was proposed. Firstly, the vibration signals of gearbox were decomposed by variational mode decomposition (VMD). Then, according to the cross-correlation criterion, the components smaller than the threshold were denoised and filtered by maximum correlation kurtosis deconvolution (MCKD), and the denoised signals were reconstructed. Finally, the fault feature sets were constructed to realize adaptive fault feature mining and intelligent fault pattern recognition based on DBN. The gearbox diagnosis in oilfield shows that the proposed method has self-adaptability, significantly improves the accuracy of fault classification, and provides a basis for ensuring the safe and reliable operation of oilfield equipments.