In this study, we examined the effects of response consensuality in syllogistic reasoning on patterns of answer change by using the two-response paradigm. Participants evaluated 24 syllogistic problems previously found to differ in consensuality, including consensually correct (CC), consensually wrong (CW), and nonconsensual (NC) items. Each problem was presented two times and participants were required to provide an initial quick answer to the first presentation, to rethink the problem, and to provide their second and final response without time limits to the second presentation. Participants reported the feeling-of-rightness (FOR) following the initial response, and the final judgment of confidence (FJC) after the final response. Following the assumptions of Koriat's (2012) Self-Consistency Model of confidence, we expected higher probability of answer change for initial nonconsensual responses than for initial consensual responses. The results showed that patterns of answer change, as well as metacognitive judgments and response times, were related to item consensus and response consensuality. Nonconsensual responses were more likely to be changed than consensual responses, and the probability of answer change correlated negatively with item consensus. Faster response times and higher FORs and FJCs were obtained for consensual and consistent responses than for nonconsensual and inconsistent responses. The obtained results indicate that answer change may in part be a consequence of random fluctuations in representation sampling, or in generating evidence that supports each of the two response options.