This paper proposes a new despeckling algorithm of SAR image based on EMD and PCA. Firstly, it approximately evaluates the noise energy in all levels of IMF according to the energy distribution model of Gaussian white noise decomposed by EMD. And then it uses PCA to decompose IMF. According to the decomposition feature of the noisy signal and the proportion of noise energy in IMF, select proper principal components to reconstruct IMF so as to further remove the IMF noise. Finally, the simulation results show that the new method can effectively delete noise, and better keep the texture details of the edges.