A visual experiment using a beam-splitter-based optical see-through augmented reality (OST-AR) setup tested the effect of the size and alignment of AR overlays with a brightness-matching task using physical cubes. Results indicate that more luminance is required when AR overlays are oversized with respect to the cubes, showing that observers discount the AR overlay to a greater extent when it is more obviously a transparent layer. This is not explained by conventional color appearance modeling but supports an AR-specific model based on foreground-background discounting. The findings and model will help determine parameters for creating convincing AR manipulation of real-world objects.