The computation of radiant fluxes from radiance measurements by satellite radiometers requires the use of a bidirectional model. The bidirectional model, in turn, depends on the scene, so that a scene identification algorithm is required. The accuracy of scene identification varies with viewing zenith angle due mainly to geometric effects of the clouds and the measurements. This paper investigates the effects of viewing zenith angle on scene identification errors. Along-track measurements from the Earth Radiation Budget Experiment (ERBE) have been analyzed to quantify the probabilities of identifying a given scene as to its level of cloudiness, as used by the ERBE data processing system. Scene identification probabilities were initially computed on the basis of a select set of pixels. These restricted scene identification probabilities were then used to compute scene identification probabilities for all pixels, including those of marginal scene type. These probabilities are found to be functions of viewing zenith angle, but are insensitive to other parameters such as solar zenith angle, relative azimuth, and latitude, except in the Sun glint region. For most cases, the scene identification probabilities change very slowly for viewing zenith angle less than 40¿, but change rapidly for viewing zenith angle greater than 50¿. With increasing viewing zenith angles, the probability increases for identifying a scene as being more cloudy than its true cloud class. These results are useful for evaluation of flux errors due to scene identification errors in the data processing. ¿ American Geophysical Union 1995