A review of 39 driver fatalities in 1990-93 cars with air bags from the National Accident Sampling System indicated most of these fatalities were due to causes unrelated to frontal air bag performance. Two-thirds occurred in side-impact or rollover crashes, in which air bag effectiveness is limited; of 15 frontal crash fatalities, 6 died of causes unrelated to the frontal impact and 5 in cars with severe intrusion. The remaining four fatalities, three of whom were unbelted, were in moderate to high severity crashes which could have been survivable; however the deploying air bags, instead of protecting, probably contributed to the fatal injuries. A similar review of 12 fatalities of unbelted drivers in cars without air bags revealed 3 could have been prevented by air bags, but 4 were in crashes that could have put them in position to be injured by the air bag. These results suggest that reducing deployment energy would improve air bag effectiveness in relatively severe crashes as well as low severity crashes, even for unbelted drivers. No examples of fatal driver injuries from air bags in low severity crashes were found.