A construction site is a harsh environment demanding entire human senses and attention, even for regular scheduled tasks. Many accidents occur on construction sites because of inability of workers to identify hazards and make timely decisions. To understand why some hazards go unseen, it is crucial to study how workers perceive the site. The objective of this research is to leverage eye-tracking technology to study workers' gazing pattern in a construction environment. A real picture from an active construction site is modified to introduce hazards and a desktop experiment is conducted, in which, subjects are asked to identify the hazards. A different group of subjects are made to make similar observations on a 2D sketch-representation of the same construction scenario. Eye-tracking data gathered from their observations is analyzed to understand when, how, and which hazards do they recognize and the pattern of recognition is studied. The results of this study will enhance our understanding on the visual factors that govern attention and help workers recognize potential hazards in a construction site. The comparison between the observation pattern in real and sketch-representation is done to assess how subjects respond to artificial images compared to real images. This comparison will test the feasibility of using virtual reality for safety training and simulations.