Air-gap discharge in oil-paper insulation is one of the main types of partial discharge (PD) in power transformer. The discharge development stage for monitoring and diagnosis of transformer potential faults is a significant area of study. The method of wavelet packet energy entropy, which is based on different frequency bands energy distribution of PD signals at different insulation states, is provided to explore the variation characteristics of the whole PD process. In this paper, air-gap discharge model is built in the simulative transformer tank that collects PD signals based on constant voltage method. This model also utilizes wavelet packet decomposition method to partition PD signal bands obtaining signal energy distribution in each frequency band, as well as total signal energy tendency along with PD development process. Wavelet packet energy entropy, which is the new PD feature parameter describing the development process, represents the order degree of PD signals which corresponds to dielectric strength. Finally, because of the cyclic change of this method, the step points of wavelet packet entropy are taken as the way to effectively divide the PD development stage.