On Aug. 8, 1980, the state and city of New York sued the Census Bureau, contending the latest census severely undercounted the city's large black and Hispanic populations.

New York City has figured it loses $26 million to $52 million a year in federal grant money because of the "undercount," said Judy Chesser, director of the city's federal office here. New York state claimed the undercount would cost it a U.S. House seat.

For the Census Bureau, the 1980 tally is history; it is deep into planning the 1990 count. But the New York suit and nearly two dozen more brought by others alleging various errors are unresolved. These leftover lawsuits illustrate the high stakes and big issues involved in the census, the nation's 21st since 1790.

Billions of dollars in federal grants to the states and localities each year for the decade of the 1990s will be handed out on the basis of census population figures. The figures will determine the number of House seats each state will have during the 1990s, and within each state, the boundaries of congressional districts and state senate, state house and local electoral districts. Businesses and government agencies all over the country will use census figures for marketing, planning, zoning and social policy.

But the problem of an undercount in 1990 is not the only problem facing the Census Bureau. The Federation for American Immigration Reform (FAIR), which works to discourage illegal immigration, also plans legal action. The bureau traditionally counts any person it can find, even those here illegally, and FAIR says it will sue to block inclusion of any illegal aliens in the 1990 census.

The bureau is also caught up in an administration battle with the Office of Management and Budget over which questions to use on the forms.

But the greatest problem may well be the massive size and complexity of the census itself. "We expect to count 250 million people" in 1990, said Susan M. Miskura, chief of the bureau's Decennial Planning Division.

The census, mandated by the Constitution to occur every 10 years, is not based on a mere sample of the population. A questionnaire is actually given or sent to every residential unit in the nation to count the number of people in every occupied house and apartment and to obtain information about their income, race, fertility patterns, marital status, housing (including number of rooms, plumbing and value) and the like. It is the only source of direct, detailed information for tiny geographic units ranging downward in size through county, town, neighborhood, block and street.

The 10-year cost of the 1990 census and the armload of tables, books and statistical compilations it produces will be $2.6 billion, double the cost of the 1980 effort.

The bureau, which has 5,000 regular employes, will hire at least 300,000 more temporarily in 1990 to conduct interviews and help process the answers -- a hiring effort almost equal to the 325,000 people who join the armed forces each year.

"To find the 300,000 to 400,000 we will use, we may have to recruit as many as 2 million people and test their math and reading and map-reading skills," said Marshall L. Turner, chief of the 1990 Census Redistricting Office.

Miskura said the bureau will mail or deliver questionnaires to all 106 million housing units in the country on or before April 1, 1990, requesting that 101 million be filled out promptly and mailed back. The other 5 million, covering homes in sparsely settled areas, will be filled out by bureau personnel on visits to the homes.

Miskura said that up to 30 million forms designated for return by mail probably will not be returned, so the bureau will send "enumerators" to these residences to see whether they are occupied and to have the forms filled out. In the end, she estimated, about 10 million to 11 million of the residences will turn out to be unoccupied.

The "dress rehearsal" for 1990 will be held on March 20 next year in St. Louis, rural Missouri and remote counties of Washington state.

Miskura said that in the 1988 test, and also, the bureau expects, in 1990, two different forms will be used: a short form with 17 basic questions that goes to five-sixths of all housing units; and a long form that contains the same 17 questions, plus numerous others, and goes to one-sixth of all households specially chosen as samples. With some questions requiring multiple answers, the long form asks about 70 questions.

OMB, which is responsible for making sure that federal forms are not too complex and overloaded, has questioned whether about half of the 70 long-form questions, involving housing, work and fertility, should be included.

The Census Bureau, some Cabinet agencies and various private users of data, such as housing groups, have reacted strongly, declaring that throwing out these questions would decimate information-gathering.

Wendy Lee Gramm, administrator of OMB's Office of Information and Regulatory Affairs, labeled as "nonsense" implications by some critics that OMB has an ideological motive for its objections -- a desire to reduce the scope of housing programs, for example, by reducing data on housing.

She also said that Census Bureau informational papers suggesting that OMB is demanding the questions be dropped are "just false -- flat false," that no formal proposal for dropping any questions had been made. All OMB is doing, Gramm said, is seeking to determine whether the same data could be adequately gathered by other means such as by using sampling techniques, instead of surveying everyone in the country.

In testimony to Congress, Gramm said that if one assigned a value of about $15 an hour to the time people spend filling out the forms, the "total cost in time and effort" for Americans to answer all the questions "will be about $450 million." She said that was a rough way of estimating the burden of paperwork.

FAIR's Roger Conner remarked caustically in an interview that with such a calculation, " 'The Cosby Show' is costing the economy $6 billion a year!"

These are minor controversies for the census-takers. "The biggest problem is to get everyone counted," said Peter Bounpane, a bureau assistant director who heads a large community outreach effort aimed "particularly {at} those historically undercounted blacks and Hispanics, and among them, younger males in big cities."

Barbara A. Bailar, associate director for statistical standards and methodology, said the bureau is aware that it does not really count everyone in the country. In 1980, it estimated the undercount at 1 percent to 2 percent, but higher for blacks and Hispanics, she said, roughly 5 percent for blacks and a bit lower for Hispanics.

Illegal aliens often avoid being counted, even though census information on individuals is confidential for 72 years and cannot be used for prosecution or even shown to other agencies. Some household residents duck the count because they live in dwellings from which they might be evicted if found.

Since the undercount rate is highest for blacks and Hispanics, jurisdictions with large minority groups feel it hurts them most, and they want the census figures adjusted to make up for any undercount.

But making an accurate adjustment is statistically difficult. The bureau will decide before the end of 1990 whether an adjustment will be included in the numbers it reports to Congress. And Congress, if wary of the results, may mandate an adjustment.