To navigate safely, pedestrians need to accurately perceive and predict other road users’ motion trajectories. Previous research has shown that the way visual information is sampled affects motion perception. Here we asked how overt attention affects time-to-arrival prediction of oncoming vehicles when viewed from a pedestrian’s point of view in a virtual road-crossing scenario. In three online experiments, we tested time-to-arrival prediction accuracies when observers pursued an approaching vehicle, fixated towards the road-crossing area, fixated towards the road close to the vehicle’s trajectory or were free to view the scene. When the observer-vehicle distance was high, participants displayed a central tendency in their predicted arrival times, indicating that vehicle speed was insufficiently taken into account when estimating its time-to-arrival. This was especially the case when participants fixated towards the road-crossing area, resulting in time-to-arrival overestimation of slow-moving vehicles and underestimation of fast-moving vehicles. The central tendency bias decreased when participants pursued the vehicle or when the eccentricity between the fixation location and the vehicle trajectory was reduced. Our results identify an unfavorable visual sampling strategy as a potential risk factor for pedestrians and suggest that overt attention is best directed towards the direction of the approaching traffic to derive accurate time-to-arrival estimates. To support pedestrian safety, we conclude that the promotion of adequate visual sampling strategies should be considered in both traffic planning and safety training measures.