A　design　of　experiment　(DOE)　methodology　based　on　numerical　simulation　is　presented　to　improve　thermal　fatigue　reliability　of　multi-row　quad　flat　non-lead　(QFN)　packages.　The　influences　of　material　properties,　structural　geometries　and　temperature　cycling　profiles　on　thermal　fatigue　life　are　evaluated.　Anand　constitutive　model　is　adopted　to　describe　the　viscoplastic　behavior　of　lead-free　solder　Sn3.0Ag0.5Cu.　The　stress　and　strain　in　solder　joints　under　temperature　cycling　are　studied　by　3D　finite　element　model.　Coffin-Manson　model　is　employed　to　predict　the　fatigue　life.　A　L27　(38)　orthogonal　array　is　built　based　on　Taguchi　method　to　figure　out　optimized　factor　combination　design　for　improving　thermal　fatigue　reliability.　The　optimized　factor　combination　design　derived　from　DOE　methodology　is　verified　by　finite　element　analysis.　Results　indicate　that　the　coefficients　of　thermal　expansion　(CTE)　of　printed　circuit　board　(PCB),　the　height　of　solder　joints　and　CTE　of　epoxy　molding　compound　(EMC)　have　critical　influence　on　thermal　fatigue　life.　The　fatigue　life　of　multi-row　QFN　package　with　original　design　is　767　cycles,　which　can　be　substantially　improved　by　5.43　times　to　4　165　cycles　after　the　optimized　factor　combination　design　based　on　the　presented　method.　©　2014　Journal　of　Mechanical　Engineering