Retrospective self-report questionnaires of negative mood states experienced in the past (e.g., the most recent two weeks) tend to be exaggerated in a negative direction relative to the average ratings given to the moods contemporaneously. The present study used three measures that decomposed mood states into their constituent elements to examine whether certain components selectively contributed to this negative bias or all components contributed to this bias equally. Fifty-three participants responded to the questionnaires via the Internet every evening for two weeks. On the final day, participants recalled and retrospectively evaluated their mood state over the previous two weeks as a whole. The results revealed that memory bias occurred selectively for negative mood states. Anxiety, depression, and helplessness were exaggerated in the global compared with the daily ratings. None of the positive mood components showed any bias in the retrospective global ratings. A regression analysis indicated that the difference in daily and global ratings for negative mood was partly explained by peak and final scores. Higher peak scores led to greater overestimation whereas final scores had smaller effects; the higher the final score was, the less participants overestimated their negative mood in the global ratings.