To　deal　with　the　blocking　artifacts　in　compressed　face　images,　a　facial　adaptive　BM3D　(Block　Matching　and　3D　filtering)　is　proposed　in　this　paper.　BM3D　is　one　of　the　typical　image　denoising　methods,　which　is　also　efficient　to　image　deblocking.　However,　the　specific　parameters　are　closely　related　to　its　performance.　On　the　observation　that　faces　are　featured　with　the　symmetric　and　similar　geometry　structures,　the　BM3D　can　be　further　improved　for　face　images　deblocking.　We　extend　the　searching　region　of　patches　according　to　the　facial　symmetry.　The　parameter　sigma　of　the　filter　is　also　adaptively　adjusted　according　to　the　location　of　eyes,　nose,　mouth　and　the　compressed　quality.　Experiments　demonstrate　that　the　proposed　method　can　achieve　better　results　in　both　the　subjective　and　objective　performance.　©　2016　IEEE.