There are a number of factors that cause motor vehicles to rollover. However, the impacts of roadway characteristics on rollover crashes have rarely been addressed in the literature. This study aims to apply a set of crash prediction models in order to estimate the number of rollovers as a function of road geometry, the environment, and traffic conditions. To this end, seven count-data models, including Poisson (PM), negative binomial (NB), heterogeneous negative binomial (HTNB), zero-inflated Poisson (ZIP), zero-inflated negative binomial (ZINB), hurdle Poisson (HP), and hurdle negative binomial (HNB) models, were developed and compared using crash data collected on 448 segments of Malaysian federal roads. The results showed that the HTNB was the best-fit model among the others to model the frequency of rollovers. The variables Light-Vehicle Traffic (LVT), horizontal curvature, access points, speed limit, and centreline median were positively associated with the crash frequency, while UnPaved Shoulder Width (UPSW) and Heavy-Vehicle Traffic (HVT) were found to have the opposite effect. The findings of this study suggest that rollovers could potentially be reduced by developing road safety countermeasures, such as access management of driveways, straightening sharp horizontal curves, widening shoulder width, better design of centreline medians, and posting lower speed limits and warning signs in areas with higher rollover tendency.