The effectiveness of airbags was determined in a recent paper by assuming that they did not affect ejection probability, and protected only in frontal, and nearfrontal, crashes; the impact-reducing effectiveness in such crashes was assumed equal to that of lap/shoulder belts. The effectiveness, estimated as 18%, thus depends on the distribution of crashes by direction of impact, and the incidence of ejection. Ejection rates and impact directions are markedly different for single-car as compared to two-car crashes, and as a function of driver age and alcohol use. This raised the possibility that the overall effectiveness estimate might reflect an average of higher effectiveness for drivers more likely to be responsible for crashes (youths, drunks, those involved in single-car crashes), and a lower effectiveness for those less likely to be responsible (non-youths, the sober, and those involved in two-car crashes). By using FARS data for 1975-1986 to calculate the distribution of fatal crashes by impact direction under various conditions, airbag effectiveness was inferred under these same conditions. Although substantial dependencies on impact direction were found, the fraction of all fatalities that were frontals remained surprisingly constant. It is concluded that effectiveness does not depend much on driver age or alcohol consumption, but is greater for two-car crashes than for single-car crashes (21% compared to 16%).