The Census Bureau said yesterday that "limited use of scientific sampling" is essential to making the year 2000 census accurate, a position that angered congressional Republicans who are insisting on an old-fashioned head count of the nation's population.

Sampling is a widely used technique of studying a portion of a population to gain knowledge that would be too costly to obtain by interviewing every member of the group. "Our plan won't make the census perfect, but it will make it a lot better," said Census Director Martha Farnsworth Riche.

But many Republicans, especially members of the House, remained highly suspicious of the technique, fearing that it will help Democrats by artificially inflating the number of minorities and city dwellers. Many GOP lawmakers also believe the Clinton White House will attempt to manipulate the census results, regardless of what might be the most scientific method.

Poor people and immigrants in urban areas, many of them children, as well as the rural poor and Native Americans are among those most often missed by traditional census methods, which involve mailings to all known households followed by repeat mailings and actual visits by census takers called enumerators.

Census results long have been fraught with politics because they are critical to determining the reapportionment of the House of Representatives and state legislatures. Billions of dollars in federal funds to states and localities are distributed based on the population information from each decennial count.

Census officials said yesterday that the latest flap represents a fundamental threat to the integrity of the counting process because sampling is so widely used in modern statistics. Statisticians say that head counts can be more inaccurate than sampling because interviewers frequently make mistakes and incorrectly count subjects. And many people go uncounted because they cannot be found -- the problem that sampling seeks to overcome.

In response to a congressional mandate, the Census Bureau yesterday revealed its plan for using sampling techniques to supplement information it attempts to get from each household in the nation. Without sampling, officials said, the 2000 census will be seriously flawed and will cost up to $800 million more than the projected $4 billion.

The sampling process proposed yesterday has two phases. First, enumerators will attempt to visit (or sample) the 22.5 million of the estimated 34 million households that are expected to fail to return mail questionnaires.

A second sample will come in the detailed interviews that enumerators are to conduct in about 25,000 neighborhoods -- or blocks as Census officials call them. Here workers will question people at an estimated 750,000 randomly selected housing units. Their objective is to compare what the interviewers find in the houses there with what was previously reported by the mailed questionnaires and follow-up interviews.

By comparing the two numbers, Census officials say they will be able to discover how much of the population was missed and produce a precise and accurate population count.

President Clinton last month vetoed legislation to provide relief to flood-ravaged states in the Upper Midwest in part because it contained a prohibition on sampling in the next census. After a public outcry, the congressional Republican leadership dropped the ban and replaced it with language calling for the report that was released yesterday. Clinton signed the new legislation.

Acting Commerce Undersecretary for Economic Affairs Lee Price said the Census Bureau was ready to work with Congress on how the 2000 census should be conducted but that Clinton is "fully committed to do an accurate census." Price said that means using sampling.

That also means, both sides said, that the issue is far from resolved. The GOP leadership is likely to use the Commerce Department's appropriations bill and other measures to spell out the conditions under which the 2000 census can be conducted. The White House and most Democrats are likely to resist.

"My party is determined not to have sampling," said Rep. Christopher Shays (R-Conn.), a co-chairman of the Congressional Census Caucus who said he believes his party's policy is "dead wrong." He warned that Clinton should be ready to veto lots of bills because Republicans "are going to put it {the sampling prohibition} on everything."

Rep. J. Dennis Hastert (R-Ill.), chairman of a House Government Reform and Oversight subcommittee that oversees the Census Bureau and an anti-sampling leader, agreed the issue is far from resolved. He rejected yesterday's Census Bureau proposal as "incomplete propaganda" and accused the bureau of withholding key information about how its sampling would work.

"The fact remains sampling is a risky scheme of dubious constitutionality," Hastert said. "While the bureau makes numerous claims about the accuracy of its sampling scheme, they still offer little in the way of substance and facts to back up their claims."

Shays said most population gains from sampling would be in the South and West, regions where Republicans are strong. As for the new Census plan, Shays added, "This report helps the cause, but I'm not sure members of my party are going to be listening."

Census officials attempted to portray their use of sampling as a scientifically accepted method that is vital to making the 2000 census accurate.

It can overcome some of the new trends that make Americans less likely than before to respond to the mail questionnaire that goes to every household and is the underpinning of the count, they said. The growing mobility of Americans and the fractionalization of the traditional family unit have made it increasingly difficult to locate the nation's population, Riche said.

"The bottom line is that under the old methods, if we can't find you, we can't count you," she said.

In the 1990 census, no group was more "undercounted" than children, Census officials said, basing their judgment on surveys conducted by the government after the 1990 head count. About half of those who were uncounted were younger than 18, officials said. About 12.2 percent of Native Americans were missed compared with an estimated 5 percent of Hispanics and 4.4 percent of African Americans, they said. Less than 1 percent of non-Hispanic whites were missed, officials said.

Without sampling, the 2000 census will miss 1.9 percent of the U.S. population. Using the sampling techniques proposed yesterday, officials said they would only miss 0.1 percent of the population.

Census officials produced detailed scientific reports by the National Research Council, an arm of the National Academy of Sciences, arguing for change in the way the census is conducted. After decades of steady improvement, the council said, the 1990 census was one of the worst conducted, failing to count 4.7 million, or 1.8 percent, of the population.

"Indeed, the 2000 census could be a statistical disaster if sampling were not allowed," said Charles L. Schultze, a former chairman of the Council of Economic Advisers who chaired a Research Council panel on the problems.

"It's important to understand that no census ever taken in the United States -- we've had 21 of them -- has ever been a complete count of everyone," said Price. "It's been the best we could do under the circumstances, and we've always missed people."

The attorneys general in the administrations of Presidents Jimmy Carter, George Bush and Clinton have upheld the validity of using sampling, Riche said. But as the Census officials left their news conference yesterday, they were greeted by an announcement that the Southern Legal Foundation Inc., an Atlanta-based law firm, would sue to block use of sampling.