Dioxin　(DXN)　is　a　highly　toxic　pollutant　emitted　during　municipal　solid　waste　incinerator　(MSWI)　process.　In　the　actual　industrial　process,　DXN　emission　concentration　is　measured　through　offline　experiment　analysis,　which　has　shortcomings　such　as　long　time　and　high　cost.　In　this　study,　a　soft-sensing　model　of　DXN　emission　concentration　was　established　by　using　MSWI　process　variables.　Random　forest　(RF)　and　gradient　boosting　decision　tree　algorithms　are　used　to　construct　ensemble　learning-based　DXN　model.　First,　RF　tree　sub-models　are　constructed　base　on　random　sampling　and　CART　regression　tree.　Then,　Gradient　boosting　decision　tree　(GBDT)　method　is　used　to　each　RF　sub-model,　in　which　one　gradient　iteration　is　performed　to　reduce　the　prediction　error.　Finally,　a　simple　average　combination　strategy　is　performed　on　these　RF　and　GBDT　based　sub-models.　Thus,　the　soft　measuring　model　of　DXN　emission　concentration　based　on　small　samples　and　high-dimensional　MSWI　process　data　is　obtained.　The　proposed　method　can　both　reduce　model　variance　and　eliminate　prediction　bias.　The　experimental　results　show　that　the　proposed　method　can　further　improve　the　prediction　performance　and　generalization　ability.