CLEAN, an iterative point-deconvolution algorithm developed originally for use in radio astronomy, was investigated as a means of restoring optical coherence tomography (OCT) images of biological tissue. The CLEAN deconvolution kernel was derived from the theoretical point-spread function of an OCT scanner, which depends on the properties of the imaging system as well as the characteristics of the scattering properties of the medium. The kernel incorporates a modification based on an inverse Wiener filter that is designed to reduce ripple artifacts in images of densely packed scatterers. Evaluation of the performance of the CLEAN algorithm was carried out on a set of images acquired with a prototype scanner with built-in specklereduction hardware. The results show the ability of the algorithm to improve the resolution of features in coherence images of scattering phantoms and living tissue. In many cases, the restored images reveal tissue morphology not evident in the unprocessed images. CLEAN tolerates speckle noise well and its performance degrades gracefully as the number of unresolvable scatterers causing mutual interference increases. Ways of coping with the long processing time required for the restorations are outlined, along with possible improvements that would permit CLEAN to take advantage of both amplitude and phase information in partially coherent interference signals.