2D Gaussian filter based on local fringe rates estimation is proposed. The method estimates the local fringe rates by using the maximum-likelihood method, and evaluates the intensity of the noise by using the local standard variance of the noise. The parameters of the filter are modified adaptively according to the above two factors, and then adaptive filtering is implemented. By using both, simulated and SIR-C/X SAR generated interferograms, the performance of the method is demonstrated and compared with the mean filter, the median filter and the Goldstein filter. The result shows that the method can filter the phase noise effectively, and preserve the details of interferometric fringes better.