Image compositing is a standard computer graphics technique used to merge independently created visual elements into a single image. However, errors in the source material or the compositing process can destroy the realism of the result. In this paper we describe a series of experiments that investigate the role that lighting errors in the source material have on the realism of composite images. We study two classes of errors: pixel errors – differences in illumination intensity or color temperature; and cue errors – differences in illumination or shadow direction. The results show that the sensitivity to errors varies with the type of error. While observers are reasonably sensitive to pixel errors that change the statistics of the foreground and background image regions, they are less sensitive to cue errors that leave the image statistics unchanged but introduce conflicting information about scene lighting conditions. These trends are modulated by the scene content. These studies represent some first steps towards developing perceptual metrics for error tolerance in image compositing that can be used to improve the fidelity and efficiency of compositing process.