Affective forecasting refers to our predictions about how pleasant or unpleasant certain future events will make us feel. People tend to overestimate the impact of future events on their emotions, which is known as impact bias. The present research focuses on whether there is impact bias in middle school students’ learning behavior and the influence of learning agency belief on affective forecasting of middle school students. To test the hypotheses, three experiments were conducted. In experiment 1, a total of 100 middle school students participated the competition of “who will be the winner”. Before the test, they predicted the pleasantness of success and unpleasantness of failure in the test. After the test, they were told that they had succeeded or failed in the test, and were asked to report their actual pleasantness of success and unpleasantness of failure in the test. Results showed that all the students overestimated both the pleasantness of success and the unpleasantness of failure. In experiment 2, a total of 240 middle school students were told to participate a fake high school entrance examination. According to their scores of learning agency belief scale, they were divided into two groups, one group is with high agency belief and the other one is with low agency belief. Before the test, they predicted the pleasantness of success and unpleasantness of failure in the test. After the test, they were told that they had succeeded or failed in the test, and were asked to report their actual pleasantness of success and unpleasantness of failure in the test. Results showed that for predicting the pleasantness of success, students with high agency beliefs showed smaller impact bias than those with low agency beliefs. However, for predicting the unpleasantness of failure, students with high agency beliefs showed bigger impact bias than those with low agency beliefs. In experiment 3, a total of 300 middle school students participated the final exam. According to their scores of learning agency belief scale, they were divided into two groups, one group is with high agency belief and the other one is with low agency belief. Before the test, they predicted the pleasantness of success and unpleasantness of failure in the test. After the test, they were told that they had succeeded or failed in the test, and were asked to report their actual pleasantness of success and unpleasantness of failure. Results showed that for predicting the pleasantness of success, students with high agency beliefs showed smaller impact bias than those with low agency beliefs. However, for predicting the unpleasantness of failure, students with high agency beliefs showed bigger impact bias than those with low agency beliefs. In sum, the present research found that middle school students overestimated the impact of future learning outcomes on their feelings. Besides, learning agency beliefs influenced middle school students’ impact bias. The present research has both important theoretical and practical implications. Limitations and further research have been discussed as well.