This study evaluated the real-world performance of crash avoidance systems (CASs) on commercial heavy vehicles using naturalistic data collection. The crash avoidance systems evaluated included:​ FCW, AEB (first generation systems) 1 , and LDW. First, the study analyzed whether CAS activations were false (no potential threat), advisory (possible threat identified), or imminent (an activation in response to a real and immediate roadway conflict). Second, the study also examined behavior in drivers’ longitudinal driving performance such as changes in activation rates, driving speeds, or driving headways. Third, the study characterized some of the environmental conditions (traffic, weather, driving maneuvers, etc.) that were associated with CAS activations. Finally, the study demonstrated how driving speeds, brake response times, and decelerations could be used to help model real world conflicts in which CAS may provide safety benefits. The output of the study may be used by CAS suppliers and truck OEMs in tailoring the performance or design of their CAS products, and by regulatory agencies in evaluating the effectiveness and overall performance of such crash avoidance systems. A total of 150 CAS-equipped tractor-trailers and their drivers from across the U.S. were recruited from seven commercial fleets to participate for up to 15 months in the field study. Data collection occurred between November 2013 and August 2015. A total of 2.9 million miles and 90,000 hours of driving data were recorded in the study. The study recorded video of the driver’s face and torso, video of the forward roadway, vehicle network data, and parametric data whenever the trucks were in motion. Approximately 6,000 CAS activations were sampled for further evaluation, including all emergency braking activations. Results include several observations on CAS and driver performance. First, false activations were observed in the data, including many stationary object alerts within the sample. Overpasses, overhead signage, roadside infrastructure (signs, etc.), and curves in the road were common causes of false stationary object alerts. Second, there were several observations about when the truck drivers’ actions triggered CAS activations versus when other vehicles triggered activations. Finally, there were observations of drivers potentially misusing controls for the lane departure warnings. The real-world situations and driver behaviors that generate activations, as well as driver behaviors in response to activations can be used by system manufacturers to improve the performance of CAS devices. False positive activations caused concern among fleet managers because drivers’ trust and use of the system is paramount to its effectiveness. This study is limited to heavy vehicle CAS systems as their performance and implementation differs from light vehicles, so results may be different on other vehicle platforms. Naturalistic methods are a valuable tool for understanding real-world performance. As CAS technologies and other automation features become more and more capable, naturalistic research will allow all interested parties to better understand the benefits and unintended consequences of real- world usage.