Crossing a road intersection, a driver must collect a certain amount of visual information from various locations. The allocation of visual attention, which allows this collection, mainly relies on top-down processes. This study focuses on three top-down factors which inuence the collection of visual information: the value of visual information for the ongoing task, their bandwidth, and the familiarity with the environment. These factors are studied according to the priority rules at intersections (Give way, Stop or Priority), the expected trac density (Lower or Higher) and the number of passages (First or Second passage). Fourteen participants were installed in an instrumented vehicle equipped with an eye-tracker. They drove during 1h45 along a 80-km long route, mainly on rural roads, which included 19 intersections. Visual attention was studied by means of the head and gaze horizontal eccentricity. Eects were found for each of the three factors, in agreement with Wickens' theoretical framework and with previous studies, despite the important variability in the data due to the experimental situation.