Drivers' speed choice on a curve depends on their perception of the safety offered by that particular geometric element. Perception is a latent variable and is difficult to measure directly from the field. This study assumed that perception is reflected in the speed and path radius adopted by drivers on curves. Preliminary analysis of the speed variability and the difference between path radius and curve radius showed that the curve direction and superimposed horizontal and vertical alignments significantly influence the driver's perception. The study suggests a clustering-based approach to deal with drivers' perception of a complex geometry. Curve direction and type of superimposed vertical curve were considered as the criteria for clustering. Based on those two criteria, the curves were clustered into eight groups. For each cluster, the geometric variables influencing the operating speed were identified through correlation analysis. The study found a significant improvement in the correlation between geometric parameters and operating speed after clustering. The results support the assumption that driver perception plays a significant role in operating speed decisions. © 2019 American Society of Civil Engineers.