Despite the advice that “don’t judge a book by its cover”, people constantly use facial appearances to form impressions or guide behaviors toward others. Its influence is so persistent that even when other types of information (e.g., behavioral information) are presented, it still impacts the judgment at hand. Does it mean the influence of faces cannot be overridden? This seems like a possibility, especially for implicit evaluations based on faces, because traditionally, implicit evaluations are argued to be much more difficult to change than explicit evaluations. Thus, in this dissertation, I include work examining the possibility of changing implicit impressions formed based on faces, in particular, facial trustworthiness. Across three papers, I demonstrate that initial implicit impressions formed on facial trustworthiness can be rapidly updated after learning new diagnostic and countervailing behavioral information. In addition, the updated implicit impressions bear potential consequences for behavioral outcomes.In the first paper, I examined whether and when face-based implicit impressions can be revised in the difficult direction—from negative to positive. Across five studies, participants who formed initial negative impressions toward an untrustworthy looking target quickly updated their impressions about that person after and only after learning new information that is extreme and diagnostic. The contents of the new information, however, does not have to strictly imply the specific trait (i.e., trust). It can also imply general valence (positive or negative). The work additionally showed that the updating process is efficient. In the second paper, I replicate the updating results with trustworthy faces and test the importance of information extremity in updating implicit evaluations. In addition, another critical factor, information reliability, is found for changing implicit impressions. The studies showed robust results of changing both trustworthy and untrustworthy faces while revealing the difficulty of eliminating the positivity associated with trustworthy faces compared with untrustworthy faces. Finally, this work demonstrates the durability of changed implicit evaluations by applying a test and retest over a three-day interval. The main focus of the third paper is to test the predictive validity of updated implicit evaluations. The updating paradigm from the first paper is adapted to induce implicit updating of impressions based on untrustworthy faces. Then, I measure whether updated implicit impressions uniquely predict trusting behaviors in the trust game. Across six studies, an overall small yet significant effect is found, suggesting the utility of implicit measures for predicting behaviors.