A novel method of interferometric synthetic aperture radar phase filtering that combines empirical mode decomposition (EMD) with Hölder exponent adjustment is presented in this letter. First, intrinsic mode functions (IMFs) of different levels are obtained by decomposing the real and imaginary parts of the noisy interferometric phase in complex formulation respectively employing EMD, which is a totally data-driven method without parameters to be selected. Then, we increase the Hölder exponents of every IMF to appropriate extent according to the features of the signal and noise contained in them to realize different filtering effects. Thus, noise can be efficiently filtered without the loss of detailed information of the interferogram. Finally, the filtered IMFs are reconstructed to form the denoised interferogram. The experiments of simulated data with various correlation coefficients and real data verify the effectiveness and adaptability of the method.