Auto　focus　is　an　important　technique　to　improve　the　automation　for　fundus　camera.　Selection　of　image　sharpness　evaluation　function　is　a　key　to　auto-focus　technique.　In　order　to　ensure　the　focusing　accuracy,　a　new　auto-focus　algorithm　is　proposed.　It　uses　the　definition　of　retinal　vessel　edge　as　sharpness　evaluation　function　and　combines　the　rough　search　with　the　fine　search　together.　The　retinal　vessel　edge　can　be　extracted　fast　and　accurately　by　the　edge　confidence　detection,　and　the　result　is　used　as　the　evaluation　function　of　the　auto-focus.　And　in　order　to　decrease　the　computation　complexity,　the　proposed　sharpness　evaluation　function　is　used　only　in　the　in-focus　area.　Experimental　results　show　that　the　approach　has　the　better　unbiasedness　than　the　traditional　gradient-based　evaluation　function,　much　more　advantages　in　focus　accuracy　and　excellent　robustness　against　noise.　©　2010　IEEE.