It began with turbulence, a little bounce and rocking, nothing particularly unusual. Rain and hail splashed the cockpit window of the Lockheed L1011 jumbo jet as it homed on the approach lights to Kansas City International Airport.

The speed gauge suddenly spun downward, from 141 knots to a dangerously slow 121.

"Glide slope. Glide slope," the computerized voice said, warning that the plane had fallen below the electronic approach path to the runway.

A more urgent warning, a command: "Pull up! Pull up! Pull up!" The rain and hail were now pounding and the runway approach lights were less distinct, but closer.

The pilot pulled back on the control column in an attempt to gain altitude, but the stick began to vibrate and a sound like a rattlesnake poised to strike filled the cockpit as the "stick shaker" alerted the crew that the plane was close to losing ability to fly.

The pilot pushed the throttles fully open and pulled farther back on the column. A jolt. The alarms stopped. Silence.

Salvatore J. Fallucco, Trans World Airlines' manager of instruction and standards, had just "crashed" the TWA L1011 flight simulator in a microburst wind shear. What started as a routine landing in a rain shower had become an impossible situation within seconds.

Microburst wind shear is a proven killer. In the past 12 years, 399 people have died in four major U.S. airline wind shear accidents. The Delta Air Lines L1011 crash in Dallas Aug. 2 has the classic fingerprints of wind shear, and if that is found to be the cause, the toll will increase by 133, to 532. In the same period, there were at least eight other major airline wind shear accidents where a jetliner was forced harshly to the ground on either takeoff or landing, but no one was killed. Many close calls have gone unreported.

The Federal Aviation Administration wind shear detection system produces more false alarms than reality, and is widely distrusted by pilots. It cannot predict microburst or other wind shears; it can only report their presence.

A new-technology detection system called terminal Doppler radar is at least four years away, and that assumes that on-again, off-again FAA research and development will move forward efficiently and rapidly.

"We call it research on a tombstone," an FAA official familiar with the program said. The Office of Management and Budget "tells us that one problem holding this up is that you guys don't have enough accidents; you can't justify an expensive detection system."

An OMB official said, "You take all the people who died in the Delta accident. More people were killed on motorcycles last week. How safe is safe?"

Wind shear's most dangerous form, the microburst, is also its most insidious. It can occur in clear, dry air, although always with thunderstorms in the area. It might last no longer than 5 minutes, perhaps as long as 20. It intensifies rapidly, then dissipates. It cannot be seen by the naked eye, only inferred from other clues not always easy to see from the cockpit.

A pilot entering a microburst at low altitude has seconds to react if he is to have any hope of saving the airplane, and he must react in ways counter to traditional pilot training.

A wind shear is a sudden shift in wind velocity or direction. In the case of a microburst, a plane can fly almost instantly from a strong head wind into a strong tail wind. That reduces the speed at which air flows over the wing, the airspeed. If the airspeed drops too low, the wing cannot support the plane.

Microbursts creating an airspeed loss of as much as 100 knots -- about 115 miles per hour -- have been measured; the average airspeed loss in a microburst is 50 knots. A plane at low altitude -- 500 feet or closer to the ground -- does not have room to adjust to such a differential.

Much of this information has been developed by Dr. John McCarthy, a meteorologist at the nonprofit National Center for Atmospheric Research in Boulder, Colo. He has headed two FAA-funded microburst-detection programs and has proven that microbursts can be detected with Doppler radar in time for pilots to be alerted.

The much-maligned "existing detection system can and should be saved," McCarthy said. "With modest cost it can be substantially improved." Doppler radar at airports "will take two to five to seven years to get on line, and between now and then there will be two or three wind shear accidents," he said.

The cost of Doppler radar for 100 major airports could range from $300 million to $700 million, depending on what is included and who makes the estimate. It would be a heavy expense at a time of governmental austerity.

Nonetheless, disasters clearly focus the governmental mind:

* After 153 people were killed in New Orleans in July 1982 when a Pan American flight was downed by a microburst after takeoff, Congress hustled through a requirement to expand the presently unreliable wind shear detection system from a handful of airports to more than 100 and ordered a National Academy of Sciences study. The academy concluded in September 1983 "that low-altitude wind variability (or wind shear) presents an infrequent but highly significant hazard to aircraft landing or taking off." It recommended improved pilot training, more research, and "immediate" FAA action to develop terminal Doppler radar.

* On May 13, 1984, after not much had happened, United Airlines Flight 663 encountered a significant shear as it was lifting off from Denver's Stapleton International Airport. The pilot, who had learned new escape techniques now taught by most airlines, was able to fly out, but could not pressurize the cabin. He returned to the airport. An antenna at the top of a 13-foot pole 1,000 feet from the end of a two-mile-long runway had punctured the airplane and was hanging from the underside of the fuselage. Shortly thereafter, the FAA found financing for McCarthy's Doppler-based pilot alerting program at Stapleton, which started running two months later. In six weeks, meteorologists issued 35 microburst advisories. Several pilots delayed takeoffs or landings and one pilot wrote that the advisory had saved his airplane, a loaded Boeing 737.

* McCarthy sought approval for a follow-on program this summer. The first proposal, for about $3 million, was rejected as too expensive. A $750,000 plan was submitted. According to an FAA document obtained by The Washington Post, "Significant budgetary restrictions were encountered . . . " An enhanced nonradar wind shear alerting system "had been funded and was considered the most cost-effective method to provide wind shear information to the tower and thus to the pilot."

* In Olive Branch, Miss., Lincoln Laboratory/Massachusetts Institute of Technology is working this summer to test various Doppler radar configurations and microburst detection programs in the weather patterns around nearby Memphis International Airport. A major effort is to determine how to suppress phony signals from buildings and airport towers so a low-altitude radar signal paints an honest picture. "We didn't know where we were going to get the money to keep that going before the Dallas crash," an FAA official said. "Now we have the money."

* Early this year, the Air Transport Association (which represents most airlines), the three major airplane manufacturers and McCarthy developed an unsolicited proposal for $800,000 in FAA funding for a pilot training package on wind shear -- how to recognize and avoid it, and also what to do if caught in it. The FAA forwarded the package to its parent Transportation Department, but it was bounced because it was not a competitive procurement. Within the past two weeks, after the Delta crash, the FAA received approval to go ahead. "The competition advocate has to get in and help us and do his thing," said Neal A. Blake, the FAA's deputy associate administrator for engineering. "It's all part of good government."

Cost-benefit analysis may become an FAA partner in making the case for terminal Doppler radar. In addition to predicting microbursts, the radar at Denver was able to tell controllers when the wind would shift and give them -- and airplanes -- advance warning on when to change runways. McCarthy said that a government analyst predicted annual airline fuel savings of at least $850,000 at Denver because of less holding and taxiing.

But for the moment, airline passengers will have to rely primarily on pilot skill to avoid or recover from wind shear. TWA and many other airlines teach their pilots a different recovery technique than the one Fallucco used in crashing the simulator in a demonstration for a reporter.

"Everything we say in our training is avoidance," Fallucco said. "We stress over and over again, there are some shears you cannot fly out of." All TWA pilots encounter an undefeatable wind shear in their simulator training.

TWA's recovery procedures were adopted in 1979. The classic technique to recover when approaching a stall is to lower the nose and go to full throttle. That's the wrong thing for a jetliner to do in a wind shear on takeoff or landing because the plane would fly right into the ground. The right thing is to go to full throttle immediately and raise the nose to the highest flyable angle.

A jet can be flown "on the stick shaker," as the pilots say, because the stick shaker is a warning, not a finality. Almost instantaneous reaction to the loss of airspeed is necessary if the plane is to recover. Raising the nose will reduce airspeed but should increase the altitude.

"You are going to have to sacrifice airspeed for altitude -- and you might not have the altitude," said Vernon L. Laursen, TWA's staff vice president for flight crew training.

The everyday passenger wants to know why pilots fly into thunderstorms in the first place. Part of the answer is that familiarity breeds contempt: most pilots have flown through rainstorms many times without incident.

With airline travel growing rapidly and airports experiencing long lines on the taxiways and holding patterns in the sky, "there are subtle messages to the pilot saying, 'Get the plane on. Get the plane off,' " McCarthy said.

Thus, the emphasis on detection. While microbursts can be detected by Doppler radar, it takes an expensive meteorologist to interpret a Doppler radar screen. That meteorologist has to be automated.

Lincoln Laboratory, in addition to perfecting the filtering systems needed to make the Doppler work at ground level, is also working to develop computer programs that will "read" the radar and flash a warning to the controller.

The FAA's technical center in Atlantic City is trying to figure what the computer should say to the controller and what the controller should say to the pilot. Pilots want to know what will happen on their flight path: What will the airspeed loss or gain be? Where will it be encountered? With that information, a pilot can make intelligent decisions about staying or going.

The FAA for years has worried about legalisms in controller-pilot communication. Donald Turnbull, FAA program manager for the Lincoln Laboratory effort, said that "one of the concerns, as you give controllers better and better weather information, is do you change his legal responsibility?"

The FAA wants all interpretation -- and thus all potential crash liability settlements -- to remain in the cockpit and with the airline. Thus controllers are discouraged from interpreting the data.

The existing low-level wind shear alert system consists of five sensors scattered around an airport and connected to a computer in the tower. If one of the five sensors reports a wind speed 15 knots different from the speed recorded by a sensor at the center of the airport, an alert goes off.

The controllers broadcast the wind speeds and direction, but provide no interpretation of what it means in terms of aircraft performance. John Blackwell, the assistant manager of plans and programs at the Memphis tower, said that "since we've had that system, to my knowledge we've never had one" pilot go around or hold as a result of an alert.

The sensors are on the airports, just above the ground. They do not measure winds on approach or takeoff paths. Microbursts can come and go between sensors and never be detected.

In New Orleans, after the crash there, the FAA expanded the system to 11 sensors, with the new sensors placed in the gaps. The expanded system has generated 24 percent more alerts than the old system would have, the FAA says.

More Doppler research is scheduled at Huntsville, Ala., next year and possibly at Denver after that to test both Doppler and the expanded existing system.

The expanded low-level system is neither a substitute for nor a forerunner to terminal Doppler, Blake said. The FAA is expanding the low-level systems at 110 airports for about $26 million and expects to complete that project by 1988.

"You're killing one and one-half to two planeloads of people a year," Blake said. "I want to save those people. Half of the microbursts occur in clear air . . . . The only thing we've got that sees them is the Doppler radar."