The signal processing based on Hilbert-Huang transform is very suitable for nonlinear and non stationary process; it can extract gear fault features effectively. In this paper we aim at the engineering need of gearbox real-time monitoring and fault diagnosis, expanding a study of JZQ250 Gear Box. We use Hilbert-Huang transform to measure gear vibration signal processing, and use the obtained instantaneous frequency Hilbert marginal spectrum as the fault feature of the gearbox fault diagnosis. Tests showed that, the marginal spectrum based on Hilbert-Huang transform can get the characteristics of fault signal frequency of the gearbox, thus it can identify the type of gearbox fault effectively and achieve early fault prediction. The characteristics of instantaneous frequency can describe the corresponding fault type better. It has a good prospect in the field of gearbox fault diagnosis.