Dioxin　(DXN)　is　a　highly　toxic　and　persistent　pollutant　discharged　from　municipal　solid　waste　incineration　(MSWI).　The　first　principal　model　of　DXN　is　difficult　to　establish　due　to　the　complex　physical　and　chemical　characteristics　of　the　incineration　process.　In　the　practical　process,　DXN　emission　concentration　is　off-line　detected　with　monthly　or　seasonal　periods.　Aim　at　such　small　sample　modeling　problem,　a　soft　measuring　method　based　on　selective　ensemble　(SEN)　least　square　support　vector　machine　(LSSVM)　for　modeling　DXN　emission　concentration　is　proposed.　At　first,　candidate　training　sub-samples　are　produced　from　original　training　samples.　Then,　different　candidate　sub-sub-models　based　on　the　same　kernel　parameter　and　regularization　parameter　are　constructed　by　using　LSSVM.　Thirdly,　ensemble　sub-models　are　selected　by　using　the　genetic　algorithm　optimization　tool　box　and　prior　knowledge.　Finally,　these　ensemble　sub-models　are　combined　by　using　partial　least　squares　algorithm　in　terms　of　reduction　con-linearity　among　different　prediction　outputs.　Simulation　results　show　effectiveness　of　the　proposed　approach　by　using　dataset　in　reference　[18].　©　2018　Technical　Committee　on　Control　Theory,　Chinese　Association　of　Automation.