This paper develops an empirical model of the relationship between road traffic accidents and traffic flows. The analysis focuses on the accident externality which is mainly determined by the difference between the marginal and average risks. The model is estimated using a new dataset which combines hourly London traffic count data from automated vehicle recorders together with police records of road accidents. The accident-flow relationship is seen to vary considerably between different road classes and geographical areas. More importantly, even having controlled for these and other differences, the accident externality is shown to vary significantly with traffic flows. In particular, while the accident externality is typically close to zero for low to moderate traffic flows, it increases substantially at high traffic flows.