Recent studies show that the phase-amplitude coupling (PAC) may play a functional role in information processing and cognition. The conventional method for PAC analysis is implemented by using Fourier-based filters, which assumes the signal is stationary and linear. However, as biological signals are nonstationary and nonlinear, this method introduces inaccuracy. To solve this problem, a new method based on Hilbert Huang Transform (HHT) is proposed in this paper, and is applied to analyze intracranial signals from two epilepsy patients. By displaying PAC comodulograms (coupling maps) in an Intrinsic Mode Function (IMF) domain instead of a traditional frequency domain, the proposed method is able to reveal regular PAC patterns in different seizure stages during seizure evolvement. Comparisons between the conventional and proposed method show the proposed one can provide not only clear but also regular PAC patterns.