With imaging depth increasing, the local contrast in the optical coherence tomography (OCT) image of object decreases, which makes it difficult to observe the internal features of object. An adaptive multi-scale Retinex (AMSR) algorithm is proposed for OCT attenuation compensation. It automatically adjust the weights of Retinex enhanced images with different scales according to the mean values of absolute differences between Retinex enhanced image with large scale and Retinex enhanced images with other small scales. This method is verified by a human fingertip OCT image and skin OCT image that AMSR can effectively compensate the light attenuation in the original OCT image, and enhance the local contrast of structure details within object. Furthermore, comparing with the original OCT image, the visual contrast measure (VCM) of the AMSR enhanced image increases 5740. This method is useful to improve the visualization of structures inside object.