Based on Hilbert-Huang transform, an adaptive phase extraction method is proposed. Signals of fringe pattern are decomposed into a series of intrinsic mode functions (IMF) by empirical mode decomposition. Criteria are presented to determine noise IMF and to identify whether mode-mixing problem exists in the corresponding noise IMF through the analysis of Hilbert spectra for each IMF. If the problem exists, new “noise” is designed adaptively according to the noise IMF and added into the original signals. The new formed signals are decomposed again. Repeat the process iteratively until the mode-mixing problem disappears. Noise IMF and background components determined in the last decomposition are subtracted from the last formed signals to get the fundamental signals, on which is performed Hilbert transform to obtain the wrapped phase. The presented method can overcome the mode-mixing problem effectively and keep detailed phase information while removing the noise and background components. The adaptability and robustness of the method is good.