The link between the electromagnetic properties of steel and its microstructure is a complex one, depending on both phase fractions and morphology. In this paper, both analytical and three-dimensional finite element (3D FEM) modelling techniques were applied to the prediction of permeability for steel with a given ferrite fraction for random ferrite/austenite distributions. Experimental measurements from a multi-frequency electromagnetic sensor on samples generated by hot isostatic pressing (HIPping) of powder mixtures were used to evaluate the analytical and FEM predictions. Theoretical treatment of the relationship between the sensor output and the effective permeability is also given; in particular, it was found that the zero crossing frequency of the real part of the inductance is approximately linearly related to the permeability for high (> 40%) ferrite percentages. The EM sensor can therefore be used to identify the samples across the full range (0-100%) of ferrite percentages using both the zero crossing frequency (> 40%) and trans-impedance (0-40%). The effect of banded (non-random) microstructures on sensor output and the prediction of the upper and lower bounds of permeability are also discussed.