By considering the correlations between cross-polarized sub-channels, the traditional multi-input multi-output (MIMO) channel models are proved to be unable to reflect the effect of antenna polarization on correlation coefficients. To solve the problem, a new modeling theory is proposed to amend the traditional model. Furthermore, the method of calculating correlation coefficients of the MIMO system with hybrid-diversity antennas in traditional channel models is proved to be an inaccurate approximation, and a new method of modeling with statistical data of MIMO systems with hybrid-diversity antennas is presented. The new model is able to replicate the correlation coefficients accurately and meet the high-precision requirements.