In this contribution, a stochastic modeling approach is proposed for characterizing the polarization status of multipath components (MPCs) in propagation channels. The 2×2 polarization matrix of each MPC is represented by the geometrical parameters of two ellipses, i.e., the ovality, tilt angle, and size of each ellipse, as well as the rotating direction of electric field intensity along the ellipse. The statistics of these parameters including correlation behaviors among them extracted from measurement data constitute the stochastic polarization model for the propagation scenario of interest. Analytical expressions are presented for the transformation from the ellipse parameters to the 2×2 polarization matrix, and vice versa. Comparing with conventional polarization models addressing merely the cross-polarization ratios, the new model provides a more complete description for the per-path polarizations in terms of the power imbalance, tilting, and polarization spread. Furthermore, based on multiple-input multiple-output channel measurement data collected with 100-MHz bandwidth and at the center frequency of 5.25 GHz, stochastic polarization models of the proposed structure are extracted for five indoor scenarios. These models are complementary to the existing geometry-based stochastic channel models for generating realistic polarization matrices for MPCs.