Temperature　is　usually　considered　as　a　fluctuation　in　near-infrared　spectral　measurement.　Chemometric　methods　were　extensively　studied　to　correct　the　effect　of　temperature　variations.　However,　temperature　can　be　considered　as　a　constructive　parameter　that　provides　detailed　chemical　information　when　systematically　changed　during　the　measurement.　Our　group　has　researched　the　relationship　between　temperature-induced　spectral　variation　(TSVC)　and　normalized　squared　temperature.　In　this　study,　we　focused　on　the　influence　of　temperature　distribution　in　calibration　set.　Multi-temperature　calibration　set　selection　(MTCS)　method　was　proposed　to　improve　the　prediction　accuracy　by　considering　the　temperature　distribution　of　calibration　samples.　Furthermore,　double-temperature　calibration　set　selection　(DTCS)　method　was　proposed　based　on　MTCS　method　and　the　relationship　between　TSVC　and　normalized　squared　temperature.　We　compare　the　prediction　performance　of　PLS　models　based　on　random　sampling　method　and　proposed　methods.　The　results　from　experimental　studies　showed　that　the　prediction　performance　was　improved　by　using　proposed　methods.　Therefore,　MTCS　method　and　DTCS　method　will　be　the　alternative　methods　to　improve　prediction　accuracy　in　near-infrared　spectral　measurement.　(C)　2018　Elsevier　B.V.　All　rights　reserved.