The aim of this paper is to introduce a novel MUSIC-like algorithm for polarized sources characterization based on a quaternion model for two-component sensor-array signal. The associated data covariance matrix is described and a comparison with the classical long-vector approach is made. We show that the use of quaternions improves the signal subspace estimation accuracy and reduces the computational burden. Additionally, the proposed algorithm presents a better resolution power for direction of arrival (DOA) estimation than the long-vector approach, for equivalent statistical performances