Abstract: Microstructures determine mechanical properties of steels, but in actual steel product process it is difficult to accurately control the microstructure to meet the requirements. General microstructure characterization methods are time consuming and results are not representative for overall quality level as only a fraction of steel sample was selected to be examined. In this paper, a macro and micro coupled 3D model was developed for nondestructively characterization of steel microstructures. For electromagnetic signals analysis, the relative permeability value computed by the micro cellular model can be used in the macro electromagnetic sensor model. The effects of different microstructure components on the relative permeability of duplex stainless steel (grain size, phase fraction, and phase distribution) were discussed. The output inductance of an electromagnetic sensor was determined by relative permeability values and can be validated experimentally. The findings indicate that the inductance value of an electromagnetic sensor at low frequency can distinguish different microstructures. This method can be applied to real-time on-line characterize steel microstructures in process of steel rolling.