Panax　notoginseng(P.notoginseng),　also　known　as　Tianqi　in　Chinese,　is　a　rare　medicinal　herb.　Panax　notoginseng　with　different　quality　grades　have　distinct　variance　of　medicinal　values.　In　this　paper,　the　near-infrared　spectroscopy　(NIRs)　measurement　technology　is　used　to　collect　the　spectral　information　of　different　grades.　The　quality　identification　model　of　Panax　notoginsen　is　established　by　multi-classification　correlation　vector　machine　(mRVM).　And　three　dimensionality　reduction　methods,　the　competitive　adaptive　reweighted　algorithm　(CARS)　and　the　firefly　algorithm　(FA)　and　successive　projections　algorithm　(SPA),　are　used　to　simplify　the　mRVM　model.　The　results　show　that　FA-mRVM,　CARS-mRVM　and　SPA-mRVM　can　eliminate　a　large　number　of　feature　variables　and　filter　out　the　effective　feature　variables.　While　FA-mRVM　has　better　performance,　with　less　numbers　of　feature　variables　and　the　highest　accuracy.　The　correct　identification　rate,　root　mean　square　error　and　average　cross-entropy　error　of　the　correction　set　and　the　prediction　set　are　100%,　0.0048,　0.0143　and　100%,　0.0046,　0.0141,　respectively.　Hence,　NIRS　combined　with　mRVM　can　be　a　useful,　rapid,　and　nondestructive　tool　to　discriminate　quality　grade　of　Panax　notoginseng.　©　2019　IEEE.