The　commercial　banks　need　identify　exceptional　client　in　their　large　number　of　customers　to　prevent　abnormal　customer＇s　risk.　In　this　paper,　four　types　of　abnormal　data　detection　method　is　introduced,　present　a　new　method　-　the　k-medoids　clustering　algorithm　combining　genetic　algorithm　to　detect　the　outlier.　Finally,　apply　the　algorithm　to　analysis　credit　data　sets,　detect　outlier　and　identify　abnormal　customer.　©　2010　IEEE.