Résumé : In an evaluative context, does the impression we think we convey to others matter, such that the more positive we think the impression conveyed is, the better we perform? Does this belief need to be accurate to perform better? We investigate the role of meta-perception and meta-accuracy in a public speaking context by asking participants to deliver a speech in front of an audience in virtual reality. Main results showed that participants' meta-perception (i.e., how positive they think the audience perceives them) was positively associated to their performance above and beyond other-perception (i.e., how the audience actually perceives them). Results also revealed that performance increased as scores of meta-perception and other-perception increased together (i.e., meta-accuracy), up to a certain threshold.