What a difference five miles an hour can make.

As the National Highway Traffic Safety Administration prepares to issue a final rule detailing how the next generation of air bags will work, the political maneuvering over what test the agency should use in the new standard has begun. Central to the skirmish between the auto industry and the federal regulators is what speed, 25 or 30 mph, the crash test should be set at to determine how protective the bag will be.

The best insight into the agency's thinking and calculations is its economic assessment of the advanced air bag proposal it issued in November. It offers several scenarios for how auto manufacturers would test--and then design--air bags. The goal is a device that would not harm certain "at risk" occupants, such as small women and children, but still offer enough protection to unbelted adult males. A final rule is due in March.

Despite the loss of 148 lives to air bags since 1990, some 4,968 lives have been saved by air bags that have deployed in crashes since the late 1980s.

The agency's economic assessment, which presents a sort of actuarial table for various kinds of air bags, shows a large difference in lives saved or lost, depending upon the speed of the test used to determine the size, shape and force of the bag.

The auto industry insists that the optimum bag should be designed around a test where a car would be crashed into a barrier at 25 mph. It says such a bag would protect male drivers, but not harm smaller occupants in a vehicle.

The agency, meanwhile, seems to be leaning toward keeping the test speed at 30 mph, which safety advocates say is necessary to produce a bag that protects unbelted drivers and passengers who are relying solely on an air bag for safety.

They say women and children would not be harmed by a redesigned 30-mph air bag because sensors can be installed to suppress the bag when there isn't enough weight in the seat, or the bag could deploy in stages instead of a single blast.

The industry's preference for a 25-mph test is linked to its experience with "depowered" air bags installed in post-1997 models. The government ordered the depowering of air bags in 1996 when it became clear that air bags had killed 148 people, most of them small-statured women and children under 12 who couldn't withstand the force of the exploding bag when they were unbelted or sitting too close to it.

The industry argues that a bag designed for a 25-mph test would approximate the kind of "depowered" bags now found in cars and trucks, which are not as big and powerful as their predecessors. Before depowering, air bags were calibrated to protect a large male, not wearing a seat belt, in a 30-mph crash.

What NHTSA economists are trying to predict in their assessment is which speed test would save, or put at risk, more lives. The numbers show broad variations in the performance of air bags designed to withstand different crash test speeds, ranging from 25 to 35 mph.

"There are give-and-takes. I don't think you can protect everybody all the time," said a NHTSA official who has worked on the calculations.

NHTSA's model used 1998 as its baseline year. It estimated that, if all cars carried air bags, 3,253 lives would have been saved in crashes. It then calculated that an additional 45 to 53 lives would be saved annually if the test required crashing a car into a barrier at 30 mph. A range is used to reflect different assumptions about the type of technology manufacturers would use to meet the rule.

If it allowed a 25-mph frontal crash test, the agency calculated that 164 to 362 lives would be lost annually. If it lets auto manufacturers continue to use a crash simulation test they have employed since 1997 to produce depowered bags, the model showed the loss of life would be even more dramatic: 397 to 687 lives.

Automakers criticize the numbers as unrealistic, saying they conflict with NHTSA's own research on the effectiveness of air bags that have been depowered to protect drivers and passengers in low-speed crashes.

"The numbers make no sense. The documents are inconsistent," said one auto industry executive, referring to NHTSA's introduction to its revised rule, which stated that there was not a loss of benefits from a slower crash test speed.

NHTSA officials said their research on depowered bags showed that many of the safety devices--though less powerful than air bags installed before 1997--still met the tougher 30-mph crash-test requirements. They explained that the estimates NHTSA arrived at on loss of life are more dire with the depowered bag test because the agency assumed in its modeling that automakers would actually begin to design air bags to meet a 25 mph standard, which could offer less protection to people involved in high-speed crashes.

The agency also assumes that the air bag problems that harmed or killed people before depowering will be eliminated to a large degree by technology that either suppresses the bag when a child is seated in front of it or inflates the bag in stages.

It said this kind of technology, if it worked properly, could eliminate all but eight of the 181 deaths estimated by the model of "at risk" occupants caused by air bags made before 1998.

But automakers insist that using a 30-mph test favors occupants who choose not to wear their seat belts. "We feel the balance should be tilted toward belted occupants, children and older adults--why disadvantage those groups for a person who only uses part of his restraint system? Why disadvantage someone who is short or young for a guy who won't wear his seat belt?" said Mitchel Scherba, director of safety integration at General Motors Corp.

NHTSA policymakers, however, said there is a stark reality that is hard to ignore when public policy is being made: Two-thirds of the 18,000 people killed annually on average in frontal crashes were not wearing a seat belt. "A lot of those people will live because the air bag is a little bit bigger," said a NHTSA analyst, referring to a design that likely would come out of a 30-mph test.