A recognition method based on variational modal decomposition and random forest is proposed for fault diagnosis of gearbox vibration signals. Firstly, a vibration signal enhancement method based on VMD and kurtosis - permutation entropy reconstruction criterion is proposed for the problem of vibration signal noise interference, and a fault feature set construction method is proposed to help extract key fault information effectively; furthermore, by carrying out the simulation experiments such as missing tooth, gear root crack, spalling, chipping tip, and so on, we obtain the fault vibration signals and the feature set, and construct a random forest-based fault type recognition model based on random forest. The results show that the recognition accuracy of the proposed method reaches more than 90% in the test set, which can provide an effective way for the recognition of gearbox faults based on vibration monitoring.