Haidt gives subjects vignettes of two cases of "harmless" taboo behavior, such as consensual adult incest and cannibalism. I want focus here on the incest case. The vignette contains no "harmful consequences" because the subjects are told that the participants in the incest have protected sex (so there is stipulated to be no risk of STI transmission or pregnancy) and suffer no harm at all . The subjects in the experiments are asked whether the case of incest is wrong or not. They tend to say yes, and when they say that it's wrong because the participants are hurt or harmed, they are reminded that it's part of the vignette that there is no harm. The subjects then tend to get flustered, saying things like "I know it's wrong, but I don't know why!" Haidt calls this "moral dumbfounding."
The error in the experimental design is that what is probably the most important moral objection to consensual adult incest-- that it's deeply harmful to appropriate family relationships, irrespective of how participants in such behavior judge that it affects their relationships-- is being ruled out by experimental fiat. Being told by the experimenter that there is no harm does not make it true, but it does make it much less likely (given obvious social dynamics among experimenters and subjects) that the subjects will cite this sort of harm as a relevant moral objection. Moreover, this is a very important error, as it means that the study is of no value for discriminating between subjects who have reasons for their judgments but are not articulating them due to the dynamic just mentioned, and those who do not have reasons for their judgments at all. By failing to discriminate between these two groups, the study fails to provide evidence for moral dumbfounding. A similar objection could be made to the cannibalism vignette.
A better design would be to give subjects a simple prompt, like "Do you think that any kind of incest- among consenting adults, for example- is ever morally acceptable?" Then, experimenters would probe subjects' responses, not by rejecting their given reasons as inappropriate, but simply by asking for further justification for each reason given. Subjects will thus eventually reach a fundamental level of moral explanation at which moral dumbfounding will become apparent, if it exists. This design avoids the confound in Haidt's study and would provide clear evidence of moral dumbfounding, an interesting phenomenon that merits a place in the psychology of moral judgment.
 All Haidt quotes are from "Moral Dumbfounding: Where Intuition Finds No Reason," coauthored with Bjorklund and Murphy. Haidt makes use of this study in arguing for a social intuitionist account of moral judgment in "The Emotional Dog and its Rational Tail: A Social Intuitionist Approach to Moral Judgment," Psychological Review, 108 (814-834).
 The vignette does not explicitly say there is no harm, but when subjects refer to possible harm they are told that there is none.