In this paper, we study the structure analysis of the Direct Sampling Method (DSM) in the threedimensional inverse electromagnetic scattering problem. Even though the DSM is well-known to the robust, fast, and efficient non-iterative type algorithm to estimate the support and shape of unknown inhomogeneities from scattered data, the inhomogeneities might not be detected in the DSM with improper test polarization. To explain the reason for the phenomenon, we carefully analyze the indicator function of DSM using the asymptotic formula of the scattered field under a small volume hypothesis of well-separated inhomogeneities. The analytic representation formula of the indicator function of DSM is presented by establishing the relationship of spherical Bessel functions, the polarization of the incident wave and the test dipole, and the polarization tensor of the inhomogeneities, which depends on their sizes, permittivity, etc. Through such theoretical results, we also show the validity of the guideline to choose the test polarization given by other work. Furthermore, we propose our method to select the polarization of the test dipole for better efficiency. With a similar path of derivation, we also introduce and analyze the indicator function of DSM in far-field configurations. Various numerical simulations with synthetic and experimental data validate our theoretical results and proposals.