The deaths of three firemen last March when their respirators apparently failed during a Texas restaurant fire have touched off a federal investigation that indicates government-certified safety equipment used by millions of workers may be unsafe.

Dr. Anthony Robbins, head of the National Institute for Occupational Safety and Health, the government's primary testing and certification agency for safety equipment, said investigators for his agency have found substantial numbers of defects in respirators, goggles, shoes, gloves and other equipment used to protect workers on the job.

Robbins, in an interview, said he has ordered a full-scale reevaluation of the NIOSH testing program.

"We're afraid a lot of this equipment that has been certified is not safe," he said. "There's a false sense of security generated by these certifications, including the ones that we approve."

According to Robbins, NIOSH safety investigators have uncovered widespread problems with safety equipment they have tested:

The investigators found that at least 10 percent of more than 250,000 self-contained respirators used by firefighters and workers handling hazardous substances are defective, Robbins said.

About 50 percent of all types of safety goggles used by workers failed to withstand NIOSH impact tests. Robbins said such impacts could send razor-sharp slivers from the glasses into the eyes of persons wearing the goggles.

When NIOSH tested the five major types of material used in safety gloves worn by workers handling the carcinogen benzene, four of the materials allowed the cancer-causing substance to seep through, Robbins said.

Robbins said the NIOSH investigations found similar problems with construction hard hats and safety shoes.

The question of whether safety devices are actually safe arose after three firemen in Lubbock, Tex., died while fighting a restaurant blaze. Investigators piecing together clues after the fire found that the diaphragms on respirators worn by all three apparently had malfunctioned.

"We started raising holy hell after that," said Michael J. Smith, director of research for the International Association of Firefighters. Smith said his group had complained to NIOSH for years about the value of respirator certifications but that nothing was done until Robbins became director of the agency seven months ago.

In May, NIOSH issued a "stop sale" order on all Airpak IIA respirators made by the Scott Aviation Corp. of Lancaster, N.Y. The firm made more than 250,000 of the devices during the last 15 years, including the respirators worn by the Lubbock firemen. NIOSH also halted sales of the Survivair Mark I, another widely used respirator.

"Last year about 160 firemen died on duty and only half of them were burned," Robbins said. "We have to suspect respirator failures in those other cases."

Edward Fierle, president of Scott Aviation, denied that his firm's respirators posed a problem for firefighters or other workers. After the Lubbock fire, he said, company investigators tested about 15,000 Airpak IIA respirators and found only a handful with diaphragm problems.

Both Robbins and Fierle acknowledged, however, that a major problem is the certification criteria used by NIOSH in testing the respirators.

According to Robbins, NIOSH does not require tests of respirators under heat or chemical corrosion conditions. The current certification requirements, he said, are mainly aimed at the use of self-contained respirators by miners.

"As far as we can tell," Robbins said, "only about 600 to 700 Scott respirators are now in use in the mines. Most are worn by firemen or workers handling things like dangerous chemicals."

Nevertheless, Robbins said, when NIOSH was formed in 1970 to act as the government's testing arm for the Labor Department's Occupational Safety and Health Administration, many of the old mining criteria for respirators were "grandfathered" into the current regulations.

Smith, of the firefighters union, was more blunt in his criticism of the respirator criteria. "We don't think they're worth a damn," he said. "They test the things at room temperature, and very few of our people work at room temperature."

Fierle said, however, that his company tests its respirators up to 200 degrees. "Above that," he said, "the average fireman will either have his ears burned or he'll run out of the place."

A broader problem, said Robbins, is NIOSH's reliance on so-called "voluntary consensus" groups to set safety equipment standards. The major such organization, he said, is the New York-based American National Standards Institute, which has helped set standards for safety goggles and respirators. $ tThe group is composed of industry, labor and NIOSH representatives. But, said Robbins, "ANSI's standard-setting committees are historically dominated by industry. Their prime function is to protect industry, not the public."

He said part of the NIOSH shakeup will include efforts by the agency to assume a greater role in setting future standards and revising existing ones. Robbins also said he was reevaluating his agency's role in future ANSI standard-setting activities.