This study investigated infants' ability to use facial expressions to predict the expressers' subsequent cooperative behavior. To explore this problem, Experiment 1 tested 10- and 14-month-olds (N = 16, respectively) by using a violation-of-expectation procedure. In the experiment, all infants were first familiarized with two models, one with a happy facial expression and the other with an angry expression. They were also familiarized with an event in which a duck puppet tried to open a box but failed. During the test phase, infants in the helping condition saw two test scenes; one in which the happy model helped the duck open the box, and the other in which the angry model helped the duck. Infants in the hindering condition saw a test scene in which the happy model hindered the duck and the other test scene in which the angry model hindered the duck. The results demonstrated that both 10- and 14-month-olds looked longer at the angry model than at the happy model in the helping condition, whereas they looked at the happy model as long as the angry model in the hindering condition. Experiment 2 tested 6-month-olds (N = 16) with a slightly modified procedure and found the same tendency as shown by 10- and 14-month-olds. These results suggest that infants as early as at 6 months do not predict that a person with an angry expression will help others. However at the same time, they do not clearly understand the incongruence between happy expressions and hindering behavior. The results were discussed by referring to a negativity bias and human environment in which infants grow up.