Driver-related factors have long been an important component in traffic safety. Studies to assess driver behavior and the related safety concerns have primarily used data that does not capture the dynamic nature of driving tasks. The widespread use of naturalistic driving data in recent years allows researchers the capability to capture real-time driver behavior and be able to infer an individual's driving style. However, current studies focus largely on at-risk safety behavior that is often incomplete (e.g., does not consider all types of at-risk safety behavior) and broadly defined regardless of the driving environment. The goal of this dissertation is to assess driver behavior in the context of the driving environment. This is accomplished using data from the second Strategic Highway Research Program (SHRP2) Naturalistic Driving Study, which includes more than 3,000 drivers on the road from 2010 to 2013. The concept of "abnormal" driving style is proposed as a complement to "normal" driving style. More specifically, the "abnormality" measures how much a driver deviates from the average driving behavior given the driving context. In this study, the average driving behavior is defined as the average of different vehicle kinematics for drivers that participated in SHRP2 and for a specific environmental context. The study thus aims to examine the association between driving "abnormality" and driver safety. Environmental factors that contribute to the formation of "normal" driving styles were identified in a systematic way through multivariate functional data clustering method and decision trees. The "abnormality" were described by a composite score as well as a set of statistical features that capture the different aspects of a driving style. Path analysis and Structural Equation Modeling method were used to reveal associations between driver safety and driving "abnormality". Results from the study provide insights into driver behavior and implications on driver safety in different environmental contexts. For example, the study showed that drivers who were more likely to crash were also more likely to have unstable lateral control on Urban Interstates. These findings can be integrated in autonomous vehicle algorithms where individual driving styles are considered. It can also provide insights on the development of new technologies to identify risky drivers and to quantify their risky levels.