Social and cognitive psychologists have conceptualized judgemental confidence (how strongly a person holds the belief that some judgement is correct) as being proportional to the amount of evidence in favour of a response. Festinger (1950) argued that there are two separate processes by which uncertainty (the inverse of confidence) can be reduced. These two processes are physical reality testing (the perceptual processing of stimulus information) and social reality testing (reliance on other people to resolve particularly ambiguous situations). However, there is surprisingly little direct evidence that uncertainty is either reduced or increased by the responses of other people. In two experimental tests (N = 74 and N = 83) it was found that disagreement increased uncertainty and agreement tended to reduce uncertainty. In a third experiment (N = 63) it was found that disagreement only increased uncertainty when stimulus information was limited, but that agreement generally reduced uncertainty. The results challenge Festinger's model of uncertainty reduction and support a self-categorization theory account.