Surface-hardened　steels　are　widely　used　in　many　industrial　components,　such　as　cam　and　gears.　Surface-hardness　(SH)　prediction　plays　an　important　role　in　assessing　the　quality　of　the　surface-hardened　steel.　In　this　paper,　an　intelligent　method　for　automatic　SH　prediction　model　establishment　is　proposed.　The　proposed　intelligent　approach　involves　three　key　steps.　Firstly,　hysteresis　loops　(HLs)　are　measured　from　the　prepared　surface-hardened　steel　rods　with　different　SHs　by　a　sensor.　Secondly,　a　binary　particle　swarm　optimization　(BPSO)　feature　selection　is　applied　to　automatically　select　optimal　feature　subset　from　the　measured　HLs　signal.　Finally,　a　single-layer　feedforward　neural　network　(FNN)　model　is　utilized　to　bridge　the　relationship　between　the　selected　features　and　SH.　The　experimental　results　show　that　the　intelligent　method　can　automatically　establish　SH　prediction　model.　The　optimal　feature　subset　based　single-layer　FNN　model　can　accurately　predict　SHs　in　surface-hardened　steel　rods　with　a　prediction　error　of　0.298　%,　indicating　that　the　intelligent　method　can　be　integrated　in　an　industrial　robot　for　SH　prediction　in　surface-hardened　steel　rod.　©　2019　IEEE.