With the aid of CERES/TRMM (Clouds and the Earth's Radiant Energy System/Tropical Rainfall Measuring Mission) data, sampling problems that may impact the construction of anisotropic models have been investigated. We show the influence of the sun-orbit geometry, which causes a lack of measurements in forward and backward scattering directions at large viewing zenith angles, and how cloud properties retrieval algorithms as obtained from the Visible Infrared Scanner (VIRS) also affect sampling. We have also focused our attention on the dependency of the cloud cover retrieval algorithm with respect to the viewing zenith angle (VZA), showing that the shape of the cloud cover probability density function changes with VZA, becoming smoother, although this change in shape is only significant for VZAs larger than around 60 degrees. From 14 days of CERES/TRMM data, SW and LW anisotropic models have been built, and a performance analysis based on the error dependency with the observing geometry has been carried out. With scene identification in terms of cloud cover, a preferred region in the flux retrieval algorithm usually appears in the region of VZAs around 55 degrees, giving the possibility to obtain flux estimates with good accuracy and simple anisotropic models.