The　conventional　approach　to　satisfying　the　fire　resistance　requirements　in　designing　reinforced　concrete　(RC)　structures　is　to　ensure　enough　concrete　cover　depth　for　the　steel　reinforcements.　The　objective　is　to　prevent　the　temperature　of　steel　reinforcements　from　rising　too　quickly.　However,　excessive　concrete　cover　depth　may　increase　the　risk　of　concrete　spalling　during　a　fire　and　lead　to　a　larger　crack　width.　Fire　is　a　natural　heat　source　that　can　activate　the　shape　memory　effect　of　iron-based　shape　memory　alloys　(Fe-SMA)　to　generate　recovery　stress.　This　study　proposes　a　new　active　approach　for　improving　the　fire　performance　of　RC　structures　based　on　the　Fe-SMA.　The　main　objective　is　to　investigate　the　feasibility　of　improving　the　fire　performance　of　RC　structures　through　the　recovery　stress　generated　by　Fe-SMA　under　fire.　The　fire　test　of　nine　Fe-SMA　RC　beams　was　conducted.　The　temperature　profile　in　the　furnace　was　similar　to　ISO834,　and　the　maximum　temperature　exceeded　800　°C.　The　flexural　behavior　of　the　concrete　beams　reinforced　with　Fe-SMA　during　and　after　the　fire　was　tested.　The　experimental　results　demonstrate　that　the　Fe-SMA　has　great　potential　in　improving　the　fire　performance　of　RC　structures.　The　mid-span　deflection　of　the　Fe-SMA　RC　beams　under　fire　is　significantly　lower　than　the　conventional　RC　beams.　The　fire　resistance　of　Fe-SMA　RC　beams　increases　from　97　min　to　133　min　at　load　level　0.3　and　from　69　min　to　103　min　at　load　level　0.5.　The　residual　deformation　of　Fe-SMA　RC　beams　reduces　from　46.3　mm　to　27.6　mm　at　load　level　0.3　and　from　34.7　mm　to　19.8　mm　at　load　level　0.5.　In　addition,　after　1　h　of　fire　exposure,　the　Fe-SMA　can　develop　permanent　prestress　after　the　beam　cools　to　room　temperature,　improving　the　mechanical　properties　after　fire.　The　residual　bearing　capacity　and　residual　stiffness　of　the　Fe-SMA　RC　beams　are　21%　and　86%–121%　higher　than　the　conventional　RC　beams,　respectively.　The　prestress　generated　by　Fe-SMA　after　1　h　of　fire　exposure　was　estimated　using　the　equivalent　section　method.　©　2024　Elsevier　Ltd