Recent　years　most　object　detection　method　tends　to　use　the　sliding　window　fashion,　which　need　to　search　the　whole　images　entirely,　causing　the　extreme　waste　of　search　resources.　In　this　paper　we　propose　an　object　detection　method　based　on　improved　Exemplar　SVMs　(IESVM).　Our　method　mainly　includes　two　steps:　(　1)　Coarse　object　detection:　we　use　a　generic　object　measure　to　find　a　region　that　may　contain　objects.　In　this　step　we　do　not　care　about　the　category　of　the　objects.　(　2)　Precise　object　detection:　we　extract　the　regions　created　in　last　step,　in　where　we　use　Exemplar　SVMs　to　finish　the　mission　of　detection.　It　is　proved　that　the　our　method　can　reduce　the　search　space　and　improve　the　accuracy　of　detection.　We　evaluate　our　IESVM　method　on　the　PASCAL　VOC　2007　dataset　and　find　that　our　method　achieves　good　results　in　the　PASCAL　object　detection　challenges.