Cloud phase recognition is important for cloud studies. The single-scattering properties of water cloud and ice cloud were studied to effectively identify them, in the light of the difference of microphysical characteristics between the two kinds of clouds, their polarized radiance was simulated by utilizing the vector radiative transfer equation based on adding-doubling method. The results indicate that the polarized radiance over a large range of scattering angles can show the distinction between water cloud and ice cloud, and can be used to recognize the cloud phase. The algorithm of cloud phase classification based on the simulations is presented and demonstrated by using the data of POLDER 0.865 μm. The results of cloud phase retrieval was compared with the phase products of MODIS and the result of cirrus detection of MODIS based on 1.38 μm. The analysis shows that the cloud phase retrieval method based on near-infrared polarimetric measurements can improve the capability of cloud phase classification through satellites.