Navigating through a road intersection is a complex manoeuvre that requires understanding the spatio-temporal relationships that exist between vehicles. Situation understanding and prediction are therefore fundamental functions for any computer-controlled safety or navigation system applied to road intersections. To interpret the situation at an intersection it is necessary to infer the intended manoeuvre of the relevant vehicles. Conventional approaches to manoeuvre prediction rely mainly on vehicle kinematics and dynamics. The contention of this paper is that contextual information in the form of topological and geometrical characteristics of the intersection can provide useful cues to understand the behaviour of a vehicle. We describe a probabilistic framework that extracts information from a digital map and uses it along with vehicle state information to estimate a driver's intended manoeuvre. The proposed approach is applicable to different types of intersections and handles uncertainty on the input information. We evaluate the performance of our approach on several real life scenarios using data recorded from real traffic.