Driving behaviors at intersections are complex. At intersections, drivers face more traffic events than elsewhere and are thus exposed to more potential errors with safety consequences. Drivers make real-time responses in a stochastic manner. This study used hidden Markov models (HMMs) to model the driving behavior of through-going vehicles on major roads at intersections. Observed vehicle movement data were used to estimate the model. A single HMM was used to cluster movements when vehicles were close to the intersection. The reestimated clustered HMMs could more accurately predict vehicle movements compared with traditional car-following models.