By analyzing a speckle noise model of optical coherence tomography (OCT) in clinical medical imaging,we propose a despeckling method for OCT images based on the local grouping principal component analysis. On the basis of the statistical characteristics of coherent images, the multiplicative noise is converted into additive noise by homomorphic filtering. By modeling pixel to be processed in the training set and its neighborhoods as a vector, we group the vectors based on the block similarity measure. Then, the principal component analysis is performed. Considering the noise interference in coherent images with the lesion, we perform the algorithm twice. Experimental results show that the proposed algorithm has better results in terms of speckle noise reduction as well as detail preservation, and satisfying objective evaluation index.