In　order　to　eliminate　redundancies　and　incompatibilities　in　samples,　extract　the　key　information,　and　so　on,　the　condition　attribute　significance　was　defined　scientifically　based　on　two　approximation　qualities　of　rough　set　according　to　the　characteristics　of　sample　information　and　the　thought　of　information　granularity.　Moreover,　a　simple　effective　method　of　attribute　reduction　was　presented,　the　main　process　was　to　determine　the　information　core,　calculate　significance　of　dispensable　condition　attribute　and　obtain　the　relative　attribute　reduction　set.　For　the　continuous　attribute　processing,　a　speedy　discretization　algorithm　was　presented　based　on　fuzzy　similar-ratio　principle　which　can　eliminate　the　fuzzy　noise.　The　typical　example　calculation　and　the　actual　application　in　oil-water　formation　recognition　system　show　that　discernment　accuracy　of　the　attribute　reduction　method　can　approach　more　than　90%,　and　the　application　effect　is　very　notable.