A　visual　attention　method　for　object　search　for　a　mobile　robot　with　an　Omni-directional　vision　system　is　presented.　Firstly,　the　explored　scene　is　separated　into　salient　and　non-salient　regions　in　the　image　and　some　salient　objects　in　above　salient　regions　are　detected　with　a　kind　of　saliency　detection　method.　And　then,　SIFT　feature　points　of　the　salient　objects　are　extracted.　The　3D　coordinates　of　the　object　feature　points,　which　are　used　to　represent　the　spatial　positions　of　the　objects,　are　computed　from　the　Omni-directional　3D　reconstruction　method.　At　last,　the　first　searching　region　is　determined　by　synthetically　considering　some　factors,　such　as　the　number　of　detected　objects　in　each　region　and　the　distance　between　the　objects　and　the　mobile　robot.　To　validate　the　effectiveness　of　the　proposed　method,　the　experiment　is　performed　in　a　realistic　indoor　scenario.　©　2011　Springer-Verlag.