As　one　of　the　prohibit　by-products　of　municipal　solid　waste　incineration　(MSWI)　process,　dioxin　(DXN)　is　difficult　to　be　on-line　measured　in　terms　of　its　multi-component　characteristic　and　complexity　production　mechanism.　Normally,　DXN　emission　concentration　is　detected　by　using　two　steps,　which　are　online　flue　gas　acquirement　with　special　instruments　in　the　factory　and　off-line　flue　gas　analysis　with　expensive　instruments　in　the　laboratory.　In　this　paper,　a　new　DXN　emission　concentration　forecasting　approach　based　on　latent　feature　extraction　and　selection　for　the　practical　MSWI　process　is　proposed.　At　first,　latent　features　of　the　high　dimensional　process　variables　are　extracted　based　on　principal　component　analysis　(PCA).　Then,　by　using　mutual　information　(MI)　and　pre-set　feature　selection　ratio,　these　latent　features　are　estimated　and　selected.　At　last,　these　selected　latent　features　are　fed　into　least-square　support　machine　vector　(LS-SVM)　model　with　super-parameter　adaptive　selection　strategy.　Simulation　results　based　on　the　practical　DXN　emission　data　of　an　industrial　MSWI　process　of　China　show　effectiveness　of　the　proposed　approach.