The operating speed model (OSM) is essential for geometric consistency analyses, but the formulation of OSMs for a complicated highway geometry is a challenging task. Highways passing through hilly terrain consist of a few geometric elements that differ from most elements because they are designed to tackle the topographical challenges. The existing practice for the development of OSMs cannot capture such distinctive elements because the data might be significantly biased to the geometric elements constituting the major part of the alignment. This paper aims to develop an OSM considering the interaction between the vehicle and complex highway geometry. The interaction between the vehicle and highway geometry can be captured through proper horizontal curve clustering. This study classifies horizontal curves into eight categories based on the turning direction of the horizontal curves (left and right) and the type of superimposed vertical alignment - hog, sag, upgrade, and downgrade. In this process, it is necessary to consider the bias in the highway geometry data resulting from the clustering. The present study develops the operating speed models considering the selection bias and heteroscedasticity in the data collected from a two-lane undivided rural road passing through hilly terrain. The findings from this study indicate that the consideration of curve clustering and selection bias resulted in improved operating speed models. Gradient at approach tangent (G1), curvature change rate (CCR and CCRS), length of the vertical curve (LV), length of approach and exit tangents (Lat and Let), and the interaction of radius with lane width (R×LW) were found to be significant in the operating speed modeling. The design consistency analysis highlights that there exists a statistically significant difference in the geometric design consistency estimated using the proposed approach compared to the ordinary least-square method. © 2021 American Society of Civil Engineers.