Terahertz time-domain spectroscopy was first used to establish a correlation with the whole-rock iron (TFe) content in different depths of the Bayan Obo protolith. Compared with element content obtained by the traditional method of X-ray fluorescence spectroscopy (XRF), a similar tendency of the absorption coefficient and refractive index is presented. Furthermore, three machine learning algorithms, namely, partial least squares regression (PLSR), random forest (RF), and multi-layer perceptron (MLP), were used to develop a quantitative analytical model for TFe content of the protolith minerals. Among the three algorithms, MLP has the highest detection accuracy, with a model coefficient of determination R 2 reaching up to 0.945. These findings demonstrate that terahertz time-domain spectroscopy can be used to rapidly quantify the TFe elemental content of protolith, providing a method of detecting the content of mineral components.