Horizontal curves on roadways are the site of numerous crashes and motorist deaths each year. Traditional methods to warn drivers of curve hazards, including static roadside signs, are sometimes ineffective at influencing driver behavior for reasons such as driver acclimation to inconsistency in posted advisory speeds. In-vehicle curve warning devices (CWDs) may be an effective alternative for reducing the number of collisions at curves. Multi-modality displays have elicited positive driver results and should be further explored. The objective of this study was to determine the most effective curve warning system using on-road performance and subjective evaluation. Two top-performing warning stimulus presentations, as determined by a recent simulator study, were tested at the Virginia Smart Road closed test highway. Both warnings exhibited auditory (speech) and visual (Heads Down Display) stimuli, however one included a throttle pushback haptic stimulus and the other did not. No on-road studies of this type of haptic stimulus, nor of CWDs, have been published to date. Forty-eight individuals, 24 age 18-25 years and 24 age >60 years, participated in the study. A 2 (Age) x 3 (Stimulus Presentation) between-subjects design was used to examine participant performance and ratings for the first "surprise" experience with the stimulus while driving on the Smart Road, and a 2x3 mixed factors design examined stimulus as a repeated measure. Participant braking reaction times, speed, and subjective evaluations were compared between stimulus presentations as well as driver age. Throttle reaction times and brake reaction times were significantly quicker, and curve entry speed significantly closer to an advisory speed for participants receiving a warning presentation than those without a warning presentation at alpha=0.05. No statistical differences between objective measures were found between the stimulus presentation with the haptic and the stimulus presentation without the haptic stimulus. Age was a significant main effect as older drivers reached more appropriate curve entry speeds than younger drivers. Driver risk-taking style was significantly related to age and to curve entry speed. During an interview, participants demonstrated higher comprehension when presented with the stimulus lacking the haptic component, and ranked this presentation higher, though ratings gathered from questionnaires were not significantly different between the two stimulus presentations. Driver comments were examined using a content analysis technique organized by design guideline topics. Discussion was presented in terms of four main research questions and recommendations toward CWD design guidelines were developed based on the objective and subjective results of this study. In addition, this research developed a foundation for further on-road testing of CWDs and other multi-modal in-vehicle warning systems. / Master of Science