This study used Witte's extended parallel process model to examine the relationships between the use of fear appeals and college students' attitudes and behavioral intentions toward global warming. A pretest-posttest quasi-experimental design was adopted. Three hundred forty-one college students from six communication courses at two universities were recruited for this experiment. The results show that when in a high-threat condition, both high- and low-efficacy messages resulted in positive attitudes and behavioral changes, while in a low-threat condition, low-efficacy messages led to negative changes in attitudes and behavioral intentions. This study suggests that when attempting to promote low carbon-emitting behaviors among college students, messages should contain both threat and efficacy information, but messages that contain low-threat and low-efficacy information should be avoided.