The　shaft　furnace　roasting　is　a　synthetic　complex　process,　its　key　technical　parameter,　magnetic-tube-recovery-rate(MTRR),　is　hard　to　be　measured　online,　so　the　optimizing　control　is　very　difficult.　An　intelligently　optimizing　control　approach　is　developed　by　combining　the　optimal　setting,　variable　prediction　and　loop　control.　The　optimal　setting　model　using　the　case-based　reasoning　provides　the　setpoints　for　the　basic　control　loops　according　to　the　real　time　prediction　of　MTRR　and　operating　conditions.　Therefore　it　achieves　the　stabilization　control　of　basic　control　loops　by　using　advanced　control　technologies.　The　proposed　approach　has　been　applied　to　the　shaft　furnace　roasting　process.　As　a　result,　the　MTRR　can　be　kept　within　optimal　ranges,　and　obvious　benefit　is　achieved.