Data on road traffic fatalities, serious casualties and slight casualties in each local authority district England and Wales were obtained for 1995-2000. District-level data were assembled for a large number of potential explanatory variables relating to population numbers and characteristics, traffic exposure, road length, curvature and junction density, land use, elevation and hilliness, and climate. Multilevel negative binomial regression models were used to identify combinations of risk factors that predicted variations in mortality and morbidity. Statistically significant explanatory variables were the expected number of casualties derived from the size and age structure of the resident population, road length and traffic counts in the district, the percentage of roads classed as minor, average cars per capita, material deprivation, the percentage of roads through urban areas and the average curvature of roads. This study demonstrates that a geographical approach to road traffic crash analysis can identify contextual associations that conventional studies of individual road sections would miss.