Considering　specialties　on　web　information　retrieval　and　analyzing　some　problems　about　the　traditional　vector　space　model,　this　paper　proposes　the　concept　of　structured　vector　space　model.　The　new　model　represents　a　web　document　as　a　logically　structured　vector,　which　contains　several　sub-vectors　related　to　relatively　independent　parts　such　as　title,　subtitle,　plain　text　and　anchor　text,　etc.　by　the　retical　analysis　and　experimental　proof.　It　can　improve　the　performance　of　the　traditional　vector　space　model　in　precision　and　recall.