Recent advances in technology have facilitated collection of continuous naturalistic data from road users while attending their daily routines. These data can be used for several analysis, including to understand which behavioral mechanisms inform drivers about the criticality of a situation and how mismatched behavioral expectations influence how a critical situation actually unfolds. But, the increasing amount of naturalistic data and its growing potential to understand the relation between road users behavior is currently lacking appropriate analysis methodologies and tools. This project aimed at understanding driver behavior mechanisms influencing the causation of crashes and will focus on analyzing and modelling road-user behavior from naturalistic data. The project starts with the development of methods for reconstruction of the pre-crash events from video and other sensor data. By understanding how road-users interact and what makes this interaction safe or unsafe in critical situations, new guideline for infrastructure design, education reinforcement, and support systems development may be derived.