Inductive　Logic　Programming　(ILP),　as　any　other　machine　learning　or　KDD　methods,　has　to　deal　with　imperfect　data　when　applied　to　real-world　problems.　Granular　Computing　(GrC)　is　a　superset　of　various　theories　(such　as　rough　sets,　fuzzy　sets　and　interval　computation)　used　to　handle　incompleteness,　uncertainty,　vagueness,　etc.　in　information　systems.　This　paper　investigates　the　feasibility　of　applying　GrC　(especially　the　rough　set　theory)　to　deal　with　imperfect　data　in　ILP.　We　list　various　kinds　of　imperfect　data　in　ILP　(including　noise　data,　too　sparse　data,　missing　data,　indiscernible　data,　and　too　strict　bias).　For　each　kind　of　imperfect　data,　we　try　to　point　out　the　resulting　problem　and　the　potential　solution　using　GrC　(or　a　particular　form　of　GrC　such　as　the　rough　set　theory).　The　presentation　includes　formalisms　suggested　by　the　authors　or　other　researchers,　as　well　as　some　general　ideas　which　may　give　rise　to　more　concrete　results　in　future　research.