Segmentation of retinal images obtained by optical coherence tomography (OCT) and retinal thickness measurement has become an important clinical diagnostic tool for many diseases in ophthalmology. However, such factors as speckle noise, low image contrast, and irregularly shaped structural features including blood vessels make it difficult to segment retinal layers accurately. An automated retinal layer segmentation method is proposed by employing block-matching and 3D filtering along with mean filtering for preprocessing and a two-step optimal search. The two-step optimal search begins with individual retinal layer segmentation by setting a variable threshold on each A-scan as initial results, which are then checked and corrected for continuity and integrity. The performance of the proposed method is tested on a set of OCT retinal images acquired from healthy people and patients. The experimental results show that the proposed method provides accurate segmentation of nine retinal layers whose mean boundary position deviation is (1.34±0.24) pixel. The method can be applied to OCT images affected by speckle noise, low image contrast, and even irregularly shaped structural features such as blood vessels.