When visual scientists talk about cues being combined in an optimal manner what they generally mean is that they are combined in the way that gives the least variability in the resulting estimate. For independent cues for the same attribute this is achieved by weighted averaging, whereby the weights only depend on the reliability of the separate cues. Consequently, the weights assigned to the cues should be completely independent of whether the cues indicate exactly the same value or are clearly in conflict, and the variability in the final estimate should be independent of the consistency between the cues. We tested the latter, somewhat counter-intuitive, prediction by comparing the reproducibility in matching a reference surface’s slant when monocular and binocular cues indicated the same slant, with the reproducibility when the cues indicated slants that differed by 15 degrees. In accordance with the counter-intuitive prediction, the variance in the matches was no larger when the cues were in conflict. This cannot be so when the conflict is so large that the cues are no longer combined by weighted averaging, but the fact that it holds for a substantial difference provides strong support for optimal cue combination in the human visual system.