The stereoscopic fusion limit denotes the largest binocular disparity for which a single fused image is perceived. Several criteria can be employed when judging whether or not a stereoscopic display is fused, and this may be a factor contributing to a discrepancy in the literature. Schor, Wood, and Ogawa (1984 Vision Research, 24, 661-665) reported that fusion limits did not change as a function of bar width, while Roumes, Plantier, Menu, and Thorpe (1997 Human Factors, 39, 359-373) reported higher fusion limits for larger stimuli than for smaller stimuli. Our investigation suggests that differing criteria between the studies could contribute to this discrepancy. In experiment 1 we measured horizontal and vertical disparity fusion limits for thin bars and for the edge of an extended surface, allowing observers to use the criterion of either diplopia or rivalry when evaluating fusion for all stimuli. Fusion limits were equal for thin bars and extended surfaces in both horizontal and vertical disparity conditions. We next measured fusion limits for a range of bar widths and instructed observers to indicate which criterion they employed on each trial. Fusion limits were constant across all stimulus widths. However, there was a sharp change in criterion from diplopia to rivalry when the angular extent of the bar width exceeded about twice the fusion limit, expressed in angular terms. We conclude that stereoscopic fusion limits do not depend on stimulus size in this context, but the criterion for fusion does. Therefore, the criterion for fusion should be clearly defined in any study measuring stereoscopic fusion limits.