The present study investigated the effects of self-assessed ability and driving experience on the ability to detect, evaluate and respond to hazards, namely, hazard perception. 86 participants took a video-based hazard perception task, using Tobbi T120 to record their eye movement. Before the experiment, participants first finished a self-reported scale and then they were divided into low ability and high ability group according to the mean score of their self-assessed ability in each driver group. A 2×2×2 mixed design was employed, with driving experience and self-assessed ability as the between-groups factors and hazard type as the within-groups factor. 18 traffic video clips, shooting from drivers’ perspective, were used and participants were asked to respond to overt hazard (visible) or covert hazard (partially invisible but urgent) quickly when they detected a potential one. Results showed there were no significant correlations between scores of drivers’ self-assessed ability and their response latency, indicating both novice and experienced drivers have bad insight into their hazard perception ability. Generally, experienced drivers reacted to overt hazards and covert hazards faster than novice drivers and self-assessed ability did affect drivers’ detection and reaction to hazards. Specifically, novice drivers, who rated their ability better, reacted to overt and covert hazards slower than their peers with low ability significantly. This suggests that novice drivers with high self-assessed ability may have a higher risk acceptance threshold, which has a negative effect on their hazard appraisal. Furthermore, no significant differences were found on hazard perception reaction time of experienced drivers with varying self-assessed ability. Eye movement data revealed that experienced drivers spent less time to detect hazards and fixated them longer than novice drivers. Compared with their peers with low ability, novice drivers with high self-assessed ability detected hazards slower and allocated less attention to them, which in turn reduced their safety margin in reaction time. This study suggests that self-assessed bias towards driving ability among different driver group should be reduced, at least properly measured, when it comes to hazard perception training and tests.