Based on the interferometric phase filtering problem of synthetic aperture radar, a modified patch-based locally optimal wiener algorithm was proposed. The proposed algorithm was the linear minimum mean square error estimator under the Gaussian additive noise condition and jointly estimated the first moment and second moment of the image, namely, mean and covariance using nonlocal means which was the state-of-art technique. When applied to interferometric phase filtering, two modifications were proposed according to the spatial variation of the noise. First, mean value, instead of median value, was used in the estimation of the noise standard deviation. Second, the number of clusters was determined adaptively according to the ratio of the maximum value to the mean value of the noise standard deviation. Experimental results on both simulation and real data show that the modified patch-based locally optimal wiener is effective and is superior to the other three algorithms.