Most operating speed studies have focused on modeling a specific percentile speed, most notably the 85th, as a function ofthe road geometrícs, This method has resulted in some drawbacks, such as the loss of information due to speed data aggregation, the inability to capture speed dispersion, and few references about the effects of the driving culture and vehicIe characteristics on the practiced speeds. Therefore, ao operating speed frontier model to improve speed prediction capabilities, is presented. The deterministic component of the model represents the maximum operating spot speed as a function ofthe local geometric features, whereas the disturbance term incIudes the nongeometric effects, such as driving behavior, type of vehicIe, and road environment. Data are collected in 88 curves and tangents of Portuguese two-lane highways located outside urban areas; approximately 18,000 free-fíow vehicIes were observed. Following ao innovative approach to operating speed modeling, the model is estimated with a stochastic frontier regression between the speeds of ali free-fíow vehicIes and the geometric features at the measurement sítes. lo addition to the maximum operating speed, the new model is capable of estimating aoy percentile speed through the cumulative function of the one-síded disturbance while avoiding speed data aggregation. Moreover, the road geometric features required to implement the model are easy to obtain either by consulting the design project or by performing on-síte measurements; this ability contributes to the model's applicability in different regions.Operating speed studies have gained relevance across the past decades since several countries started to consider the predicted dri ving speed as an input to the definition of roadway geometric standards in the guidelines for road designo In several studies the research community, public authorities, and road operators have developed the prediction of operating speed and evaluated the effects of different factors on the speed, such as road geometry and functional classification, roadside interference, traftic, speed lirnits, and weather conditions. These studies produced a large number of tools for speed modeling (1, 2) and design consistency evaluation (3, 4) that are used by practitioners worldwide.The AASHTO Green Book recognizes the 85th percentile of the speed distribution as the most commonly used operating speed measure (5). However, in Transportation Research Circular E-CI5I (1), it is pointed out that most regression models estimate only a specific percentile speed, which is one of the main deficiencies in speed modeling. Tarris et alo reported that the loss of information due to speed data aggregation reduces the total variability and the nature of the variability associated with the regression function; this loss may bias the influence of road geometrics (6). Tarris et alo propose that modeling the entire free-f1ow speed distribution may help to overcome the problem. Figueroa Medina and Tarko developed speed models for different percent...