The purpose of this study was to estimate the size of the influence of ambient light level on fatal pedestrian and vehicle crashes in three scenarios. The scenarios were: fatal pedestrian crashes at intersections, fatal pedestrian crashes on dark rural roads, and fatal single-vehicle run-off-road crashes on dark, curved roads. Each scenario's sensitivity to light level was evaluated by comparing the number of fatal crashes across changes to and from daylight saving time, within daily time periods in which an abrupt change in light level occurs relative to official clock time. The analyses included 11 years of fatal crashes in the United States, between 1987 and 1997. Scenarios involving pedestrians were most sensitive to light level, in some cases showing up to seven times more risk at night over daytime. In contrast, single-vehicle run-off-road crashes showed little difference between light and dark time periods, suggesting factors other than light level play the dominant role in these crashes. These results are discussed in the context of the possible safety improvements offered by new developments in adaptive vehicle headlighting.