Water droplet size, or effective radius, of water clouds can be retrieved using measurements of the total reflectance in near-infrared channels (bi-spectral method) or the linearly polarized radiance of the cloud rainbow (polarimetric method). The retrieved droplet sizes from the bi-spectral method can be significantly larger than those from the polarimetric method. Droplet size vertical heterogeneity is considered as a likely cause of this difference. In this radiative transfer theoretical study, we find that partially melting droplets in mixed-phase clouds can be another cause of this difference. The theoretical study suggests that, when the clouds are dominated by large particles, the existence of partially melting droplets could cause the polarimetric method to underestimate the particle size; but for moderate-sized particles, the polarimetric method should be able to retrieve accurate particle size no matter if the clouds consist of partially melting droplets or not. However, the droplet effective radius of partially melting droplets retrieved by the bi-spectral method can be overestimated if the infrared channel of the bi-spectral method is at 1.6 μm, but is relatively insensitive to the presence of partially melted spherical droplets when the infrared channel is at 2.1 μm. The retrieved droplet size difference between droplet sizes from 1.6 μm and 2.1 μm channels can be used for detecting partially melting droplets.