Standard cloud remote sensing techniques rely on two basic assumptions: First, clouds are assumed to be plane-parallel and homogeneous within each satellite pixel. Second, pixels are assumed independent and the net horizontal radiative transport between pixels is neglected. These assumptions cause considerable uncertainty and bias in the retrieval of cloud properties, which depend on the sensors spatial resolution as well as the illumination and observation geometry. The errors are quantified for several typical sensor settings. The basis of the investigation is a data set of high-resolution three-dimensional cloud property distributions of marine stratocumulus obtained from airborne radiance observations. For this predefined cloud data the sensor signals are simulated using a three-dimensional Monte Carlo radiative transfer model. Cloud properties (optical thickness, effective radius) are retrieved for the simulated observations using a two-channel retrieval, which are then compared to the given cloud data. For the retrieval of the optical thickness the main findings are large uncertainty of individual pixel values occurs for a high spatial resolution (e.g., airborne sensors) due to nonnegligible horizontal photon transport at this pixel size. For the typical polar-orbiting and geostationary sensor settings the neglect of subpixel inhomogeneity takes effect as well. Nonetheless, biases are generally small within ¿5%, if pixels are overcast. If this is not guaranteed, the bias grows rapidly, for example, to typical underestimations of 20% and more for a geostationary sensor. For the retrieval of effective radius values are generally found to be about 5% larger than expected for idealized homogeneous plane-parallel cloud conditions.