The earth’s radiation budget is sensitive to changes in the microphysical properties of low-level stratiform clouds. Their extensive coverage can significantly reduce the solar energy absorbed by the earth system. An estimate of reducing the global-mean droplet effective radius (DER) of low-level clouds by ~2 mm while keeping column water constant would balance the warming due to a CO2 doubling in the atmosphere (Slingo 1990). Thus, accurate determination of the DER of low-level clouds is essential in radiative transfer and climate modelling studies. Satellite observations have been the primary source for routinely obtaining DER information on large scales. However, due to the ubiquitous presence of thin ice clouds, identifying uniform low-level clouds from satellite imagery is a major challenge. Threshold methods are commonly applied to satellite observations at visible and infrared channels for cloud type classification. The method is not optimal for discriminating the contamination of upper-level thin ice clouds that reside above the low-level thick clouds because the thin ice clouds have weak effects on the reflectivity at visible wavelengths and emission at infrared wavelengths. However, their influence on the determination of DER for low-level clouds may be rather significant, especially when measurements at a strong absorption band such as 3.75 mm (or 1.65 mm) are used. Measurements at these wavelengths are dictated primarily by the cloud particles encountered first. Since high-level thin clouds contain mostly large ice crystals, the contamination is likely to cause overestimation of DER. This study evaluates the effect by comparing the DER retrievals with and without the presence of the upper-level thin clouds overlying a uniform low-level cloud layer. The spatial coherence method is used to identify uniform low-level clouds and a split-window brightness temperature difference method (BTDM) is used for identifying pixels that are contaminated by upper-level thin ice clouds.