Previous studies have found that the durations of fear-relevant stimuli were overestimated compared to those of neutral stimuli, even when the fear-relevant stimuli were only anticipated. The current study aimed to investigate the effect of the predictability of fear-relevant stimuli on sub-second temporal estimations. In Experiments 1a and 1b, a randomized design was employed to render the emotional valence of each trial unpredictable. In Experiments 2a and 2b, we incorporated a block design and a cueing paradigm, respectively, to render the emotional stimuli predictable. Compared with the neutral condition, the estimated blank interval was judged as being shorter under the unpredictable fear-relevant condition, while it was judged as being longer under the predictable fear-relevant condition. In other words, the unpredictable and predictable fear-relevant stimuli led to opposing temporal distortions. These results demonstrated that emotions modulate interval perception during different time processing stages.