Lecture videos are the major components in MOOCs. It is common for MOOC analytics researchers to model video behaviors in order to identify at-risk students. Much of the work emphasized prediction. However, we have little empirical understanding about these video interactions, especially at the click-level. For example, what kind of video interactions may indicate a student has experienced difficulty? To what extent can video interactions tell us about perceived video difficulty? In this paper, we present a video interaction analysis to provide empirical evidence about this issue. We find out that speed decreases, frequent and long pauses, infrequent seeks with high amount of skipping and re-watching indicate higher level of video difficulty. MOOC practitioners and instructors may use the insights to provide students with proper support to enhance the learning experience.