A　stochastic　frontier　approach　(SFA)　model　of　translog　production　function　was　constructed　to　analyze　the　growth　effect　of　agricultural　production　factors　on　grain　production　in　China.　Under　the　condition　of　unchanged　cultivated　land,　the　agricultural　labor,　capital,　and　water　were　regarded　as　input　elements　of　the　agricultural　production　function.　The　maximum　likelihood　estimation　(MLE)　method　was　used　to　analyze　the　technical　efficiency,　output　elasticity,　substitution　elasticity,　and　relative　variability　of　grain　production　in　China　from　2004　to　2018.　The　results　showed　that:　(1)　For　the　technical　efficiency　and　output　elasticity　of　the　input　factors　of　grain　production,　there　were　significant　differences　in　different　provinces.　For　example,　the　water　resource　was　insufficient　in　Beijing　and　Shanghai,　but　the　output　elasticity　of　water　was　high.　Heilongjiang　was　rich　in　water　and　had　high　technical　efficiency.　For　Xinjiang,　water　was　sufficient,　but　its　output　elasticity　was　deficient　and　the　technical　efficiency　didn＇t　increase.　(2)　The　overall　technical　efficiency　level　was　relatively　low　and　was　still　declining　year　by　year;　the　output　elasticity　of　water　was　much　greater　than　that　of　capital.　There　was　still　great　potential　for　grain　growth.　(3)　Optimizing　resource　allocation　and　controlling　the　appropriate　ratio　of　input　factors　to　develop　grain　production　could　achieve　the　maximum　benefits.　Finally,　according　to　the　empirical　results,　this　paper　put　forward　some　practical　policy　suggestions　for　optimizing　the　allocation　of　input　factors,　especially　water　and　capital,　which　can　ultimately　improve　agricultural　productivity　by　improving　technical　efficiency.