k-Anonymity　is　a　well-researched　mechanism　for　protecting　private　information　released　in　Web.　It　requires　that　each　tuple　of　a　public　released　table　must　be　indistinguishable　from　at　least　other　k-1　tuples.　Subject　to　this　constraint,　how　to　release　data　as　useful　as　possible　is　challenge.　Most　previous　works　try　to　develop　flexible　anonymization　method　to　reduce　information　loss,　however,　utility　of　released　data　is　ignored.　This　paper　studies　utility-based　k-anonymization.　We　first　analyze　deficiencies　of　previous　global　recoding　model　and　the　state-of-the-art　local　recoding　model　from　utility　view.　Best　of　our　knowledge,　this　is　the　first　paper　to　evaluate　the　two　models　from　such　a　view.　Effectively　combing　both　global　and　local　recoding,　we　then　propose　a　hybrid　algorithm　for　utility　based　k-anonymization.　The　algorithm　greedily　partitions　original　table　into　non-overlapping　sub-tables　in　global　recoding　phase,　and　then　employ　local　recoding　to　go　on　divide　each　sub-table　into　smaller　ones　if　possible.　Experiments　on　famous　adult　data　set　show　the　utility　and　also　information　loss　advantage　of　our　algorithm　towards　the　advanced　algorithms　proposed　in　recent　related　literatures.