Abstract: Metal magnetic memory detection technology has been widely studied because it can identify damage to ferromagnetic components quickly and conveniently, and it is considered to have the ability to identify hidden damage. To promote the application of metal magnetic memory technology in the damage detection of a bridge steel box beam, a static bending test on the steel box beam of the bridge was performed, and the magnetic signal distribution of the upper flange with the most severe deformation was extracted. The quantitative relationship between the stress in the damaged region and magnetic signal or magnetic signal gradient was established, and an approach for characterizing the stress and damage state of the steel beam was proposed using the magnetic field gradient index. The results show that the magnetic signal curve of the upper flange is opposite to that of the stress change form, and the magnetic signal curve reverses to a negative value after entering the plastic state and increases with the stress change speed, so the component can be judged to enter the plastic state and soon be damaged. The maximum value of the magnetic field gradient curve appears in the position with the most severe damage, and with the increase in the load, the maximum value point of the magnetic gradient constantly moves to the middle of the steel beam; thus the early warning of the failure state can be conducted. The relationship curve between the magnetic field gradient and stress can obviously distinguish the entire stress process of the component, which includes four states: initial, yield, plasticity, and damage. The stress state and damage state of components can be characterized using the magnetic field gradient index. This study can provide a reference and basis for the application of metal magnetic memory detection technology in the quantitative assessment and early warning of the damage status of bridge steel box beams.