A new National Academy of Sciences report yesterday concluded that the health threat posed by low-level nuclear radiation is considerably less than most scientists have previously calculated.

The report immediately came under attack from a dissenting academy member -- the author of a radiation report made only last year -- saying that the risk, while low, is greater than yesterday's report says and "far from negligible."

Yesterday's report said that the average person is exposed to only a fifth of a "rad" of radiation per year, half of it from natural sources in the earth and the air, half from medical and dental X-rays.

Exposure to one rad yearly for a lifetime -- far more than almost anyone is exposed to -- would cause a 3 to 8 percent increase in deaths from cancer or about 5,000 to 13,000 excess deaths among a million persons, the report said. About 164,000 cancer deaths would normally be expected in a million people, it said.

Even such an increase would be detectable only "by statistical means," taht is, by counting deaths in the entire population, the report added. "The cancer of any given person cannot be attributed with certainty" to low-level radiation, and "in general, the smaller the radiation dose the smaller the likelihood that radiation was the cause."

The report was the latest and, academy officials hoped, the last of a current series of "BEIR Reports" from the science academy's Committee on the Biological Effects of Ionizing Radiation.

A report last year, written by a committee headed by Dr. Edward P. Radford of the University of Pittsburgh, also placed low-level radiation -- nuclear radiation, X-radiation and similar "ionizing" or cell-affecting rays -- low on the scale of human risks, and much lower than a first academy BEIR report estimated in 1972.

The 1979 report triggered a revolt by some committee members, who claimed that it nonetheless overestimated low-level radiation's effects "by a factor of 10."

Yesterday's version, written by a new committee that excluded Radford and included some of the dissenters, downgraded last year's mortality estimates only slightly.

But Radford wrote a dissenting view attached to the new document. He said in an interview yesterday that the new report, through its mathematics and by stressing cancer deaths rather than cancer cases, may understate real cancer risk "by about 2 1/2 times."

In particular, he said, an academy statement summarizing the report fails to state that women have a much greater lifetime risk than men. Males, he said, who have 260 to 860 cancer cases per million per excess rad in a lifetime, while females will have 550 to 1,623 excess cases per million.

Emphasizing cancer death rates rather than cancer case rates falsely downgrades cancer as "a social problem," he maintained. "When you look at mortality data, cancer is much underreported," he said. This is in the main because of the situation in two cancers -- thyroid cancer, which is fairly treatable, and breast cancer, where new treatments have resulted In longer life spans so the breast cancer victims more often die of other causes.

Radford, though a critic of the new report, does not oppose most useful sources of low-level radiation. He favors nuclear power, for example, "if we take into account the lessons of Three Mile Island."

He said in an interview last year, one he endorsed yesterday, that for the average person, low-level radiation does increase the risk of cancer, but not by much. "It is a risk," he said, "but it is not the end of the world."

There are scientists who take a dimmer view of low-level radiation, but they are in a minority among those who have studied the subject. One member of the new academy panel, Dr. Harald Rossi of Columbia University, believes that the committee should have made an even lower estimate of cancer risk.

The new panel -- named by academy president Philip Handler to try to resolve last year's differences -- presented not one rigid risk estimate but a range based on different mathematical models that reflect the uncertainties and lack of precise knowledge scientists face in estimating such risks.

The new panel relied primarily on a new "linear-quadratic" model, a curve indicating that the cancer risk may drop to a greater extent than the risk calculated by using the 1979 mathematical model. This was a pure "linear," or straight-line model, in which there is no threshold for cancer risk and some risk at every exposure down to zero, in proportion to the amount exposure.