Risk recognition training is an important training in construction companies to avoid work-related accidents. The existing risk recognition training is usually conducted on an illustration including explicit risks. However, it is difficult to express dynamic scenarios and surroundings of the scene due to the still illustration. In this study, a video based risk recognition training tool with an eye tracking device has been developed. We applied the tool to discover differences of attentions between veterans and we found out that veterans to do risk recognition based on meta-knowledge (risk recognition process) and domain-knowledge (individual knowledge of construction work).