Interferometric phase denoising is a crucial step of interferometric synthetic aperture radar processing flow because it has significant influence on the following steps. Traditional interferometric phase denoising algorithms have a similar drawback that they will wipe off some texture details in phase images while denoising. Nonlocal (NL) means filter, in contrast, can reach a balance between denoising and texture preserving because it utilizes the feature of recursive structures in the whole image. Taking the characteristics of interferometric phase image into consideration, this letter proposes a modified NL means filter algorithm for phase denoising. Moreover, in order to preserve texture to the biggest extent, an iterative algorithm is invented. In the end, experiments on synthetic and real data validate that this algorithm outperforms other traditional denoising methods.