The most detailed government study of potential consequences of accidents at atomic power plants has concluded that the worst-case death toll could exceed 100,000 persons and damage could top $300 billion at certain locations.

The new estimates, which greatly surpass the worst-case estimate of 3,300 early deaths and $14 billion in property damage contained in the Nuclear Regulatory Commission's last safety study, which was issued in 1975, come from a study conducted for the NRC by Sandia National Laboratories.

A sophisticated computer model, called CRAC2, used the most comprehensive meteorological, population and economic data ever assembled to calculate for the first time a wide variety of possible accident consequences for each of the 80 sites in the United States where atomic power plants are operating or are under construction.

The worst-case scenario postulated in the study is based on what the commission calls a "Group 1" accident, one involving severe core damage, melting of uranium fuel, essential failure of all safety systems and a major breach of the reactor's containment resulting in a large release of radioactivity into the atmosphere.

The NRC staff has estimated the probability of such an accident -- which would be far more severe than the 1979 Three Mile Island accident, which remains the nation's most serious -- as 1 in 100,000 reactor years. Given the number of atomic power plants operating or planned, this would mean there is approximately a 2 percent chance of such an accident occurring in the United States before the year 2000.

Even if such an accident occurred, the computer model determined several weather and evacuation scenarios that could result in death and damage tolls far lower than the worst-case estimate. For some atomic plants, the study found scenarios in which a "Group 1" accident would produce relatively few early deaths and only hundreds of latent cancer deaths.

For such an accident to produce the worst-case death and damage toll calculated by the computer study, the draft report said a "Group 1" accident generally would have to be followed by a "rainout of the radioactive plume cloud onto a population center." The report termed that combination of events "improbable."

Nevertheless, many experts view the NRC's efforts to estimate the probability of accidents, particularly when dealing with complex facilities such as nuclear reactors, as an inexact art at best. Several members of the NRC's Advisory Committee on Reactor Safeguards recently termed use of such a system to predict likelihood of a core meltdown as a "sham."

Whatever the likelihood of a worst-case accident, the study makes clear that the potential death and damage toll at some atomic power plants is far greater than the NRC has acknowledged.

The computer study found that the highest death toll would occur if the worst-case accident took place at the Salem, N.J., nuclear power plant on the Delaware River. Such an accident, according to the study, could result in 102,000 "early" deaths, within a year of the accident.

The area in which deaths would occur from a worst-case accident at the Salem plant, according to the study, could include Wilmington 20 miles north of the plant.

The study found the greatest damage would occur if a worst-case accident took place at the Indian Point 3 reactor on the Hudson River 25 miles north of New York City. Such an accident, the study found, could result in $314 billion in damage.

The area in which early deaths would occur, the study found, would extend only 17 1/2 miles from Indian Point, but radiation-related injuries could occur within a 50-mile radius of the reactor. More than 17 million people live within 50 miles of the Indian Point plant. As many as56,600 early deaths could occur in a worst-case accident at that reactor, the study found.

The study found that the consequences of a worst-case accident would vary widely at the three atomic power stations serving the Washington metropolitan area.

A worst-case accident at Virginia Electric and Power Co.'s Surry plants on the James River 17 miles northwest of Newport News could result in 42,900 early deaths, according to the study.

A worst-case accident at Vepco's North Anna station on Lake Anna 40 miles northwest of Richmond could result in 2,290 early deaths, according to the study.

A worst-case accident at the Baltimore Gas & Electric Co.'s Calvert Cliffs plant on the Chesapeake Bay 40 miles south of Annapolis could result in 7,090 early deaths, according to the study.

While the draft report on the two-year Sandia study notes that it examined the possibility of worst-case accidents and accidents of lesser magnitude that occurred under weather conditions that diminished the consequences, the version given to The Washington Post does not contain the worst-case figures. The NRC is expected to release this version.

Rep. Edward J. Markey (D-Mass.), chairman of the House Interior and Insular Affairs subcommittee on oversight and investigations, obtained the full computer results submitted by Sandia to the NRC and a companion Sandia report on financial consequences of nuclear accidents. Markey made this information available to The Post.

"Perhaps the authors of the study thought the high-consequence figures would be misinterpreted, but the NRC is misleading the public when it is not honest about what the worst-case possibility is," Markey said yesterday.

The computer study, initiated in 1980 as part of an effort to develop new criteria for locating future atomic power stations, demonstrated for the first time how greatly the consequences of an accident could vary depending on winds, rain, emergency response planning and population distribution around existing plants.

The most important factor in determining the toll from a nuclear accident, the draft report of the study makes clear, is whether persons living around the plant are clustered in towns and whether the plant is within 25 miles of a major city.

"Irrespective of size, population centers beyond 25 miles do not contribute to early fatalities," the report said, "because, even for unfavorable meteorological conditions, plume concentrations of radiation fall below all early fatality thresholds before that distance."

In some cases, the report said, the maximum distance from the plant at which there would be early fatalities might be as little as 13 miles depending on meteorological conditions.

The report said that, if there are towns or cities located 10 to 20 miles from atomic plants, the number of early fatalities in a worst-case accident might "increase substantially" and that the number of deaths even under more favorable conditions could be twice as high as it would be if the population were more spread out.

"I think this raises the question," Markey said, "of whether some plants in densely populated areas should be allowed to continue to operate their full expected life. Certainly, no new plants should be constructed this close to cities, and those in the pipeline need to be seriously reevaluated."

A second major finding, according to the study, was that summary evacuation of residents from within a 10-mile radius of atomic plants could significantly reduce the number of early deaths resulting from most accidents.

However, current evacuation plans may not appreciably affect early death-toll estimates for most worst-case accidents, the report said, since these involve "rainout of radioactivity from the plume onto cities located more than 10 miles from the reactor" and federal laws now require evacuation zones of only 10 miles around each reactor site.

"The fact of the matter is that few areas of the country are adequately prepared to cope with a serious nuclear accident," Markey said. "Evacuation zones need to be enlarged in some cases, and other measures such as sheltering and public education are desperately needed."

Financial consequences of a worst-case accident also far exceeded those derived by earlier studies.

Markey suggested that this calls into sharp question the adequacy of the Price-Anderson Act of 1957, which even though amended still limits to $560 million the nuclear industry's liability for damages in the event of an "extraordinary nuclear occurrence."

"Clearly, the present level of nuclear insurance is woefully inadequate in view of the new consequence figures," Markey said, adding that he intends to see that the next Congress debates "whether the act should be abolished or the damage liability limit raised."

Sandia National Laboratories, which performed the study for the NRC, is owned by the U.S. Department of Energy and is operated under contract by AT&T.