Thermal　error　which　has　been　widely　studied　in　cutting　machine　tools,　was　ignored　in　the　EDM　machines　in　most　cases,　since　there　is　usually　no　high-speed　rotation　for　spindles.　However,　for　large　die-sinking　EDM　machines,　due　to　heavy　load　of　drive　system　and　long　processing　cycle　of　large　aeronautical　parts,　thermal　error　induced　by　jump　motion　has　seriously　impaired　the　machining　accuracy　and　gradually　been　recognized.　In　this　paper,　the　dynamic　thermal　behavior　of　spindle　induced　by　periodic　jump　motions　in　large　precision　die-sinking　EDM　machine　was　studied　for　the　first　time.　Noted　that　the　Z-axis　base　and　column　show　obvious　temperature　rise　and　the　thermal　error　in　Y　direction　is　the　largest,　which　is　about　6.5　and　5　times　compared　with　that　in　X　and　Z　directions.　Based　on　this,　an　efficient　thermal　error　prediction　model　was　presented.　Thermal　sensitive　points　were　picked　out　through　fuzzy　clustering　and　correlation　theory,　taken　as　inputs　of　radial　basis　function　(RBF)　neural　network　to　guarantee　the　accuracy.　As　a　result,　the　prediction　accuracy　in　X,　Y　and　Z　directions　are　95.2　%,　92.5　%　and　94.4　%,　respectively.　Finally,　the　effect　of　jump　period　on　spindle　thermal　behavior　was　investigated,　and　suggestions　for　optimizing　jump　motion　parameters　were　proposed　to　further　improve　the　machining　accuracy　of　large　EDM　machines.