The extreme value of interference (EVI) algorithm is a very fast and efficient method for the fringe pattern phase demodulation. It requires only two arbitrarily phase-shifted frames in which the phase shift between interferograms can be determined by searching the maximum and the minimum of the normalized interference patterns, then the measured phase is obtained by an arctangent function. Compared with other two-frame demodulation algorithms, the EVI algorithm has great advantages. Firstly, the EVI algorithm is simple and the calculation speed is fast. Secondly and more importantly, it works very well even if the number of fringes of the interferogram is less than one. However, to make this method work, the fringe should be normalized in advance, which is sometimes not a satisfactory requirement. The effects of uneven background terms, modulation amplitude variations, and random noise in the fringe pattern will make the normalization of the fringes extremely complex. Therefore, by employing the HilbertHuang transform (HHT) based prefiltering in this paper, the background intensities and modulation amplitudes of the two interferograms are suppressed and normalized respectively. Then, phase demodulation is implemented using the EVI method. Because of the HHT process, the demodulation result is greatly improved in plenty of situations. Both simulation and experimental studies have shown that the proposed improved method makes it easier to determine the phase distribution with high precision even under complex conditions.