Several sensors are dedicated to cloud observations. Among them, POLDER/PARASOL measures total and polarized visible radiances in up to 16 directions. As for other sensors, the POLDER cloud retrieval algorithm is based on the assumption that clouds are plan-parallel, homogeneous and infinite. To assess the cloud heterogeneities impacts on POLDER radiances and thus on retrieved cloud parameters, we developed a tri-dimensional radiative transfer model called 3DMCPOL allowing the computation of total and polarized radiances of 3D cloud fields. The input cloud properties are simulated with a model called 3Dcloud, which is based on a simplified dynamic/thermodynamic scheme to get cloud characteristic shape coupled with a Fourier stochastic approach to enforce cloud scale invariance. The POLDER algorithm is next applied to the simulated radiances to assess the 3D errors on the retrieved cloud parameters.