The Census Bureau recently released a broadbrush estimate of what has happened to the U.S. public in the last decade. To the companies that make a living burrowing through the minutiae of the millions, the report is relatively meaningless.

The reason is that the Census Bureau gave an overview of the states and metropolitan areas with over a million people. What the bureau hasn't released are the computer tapes with estimates of how much money people make, or what they do for a living in a particular block or zip code.

"Our clients aren't trying to figure out what state to be in, they want to know where to be in town," said Patsy Bailey Allard, census liaison for National Planning Data Corp., a census data extraction firm in Ithaca, N.Y. So the data mining companies provide their own best estimates of where the buying power is.

Only a very small portion of the census data is published. The "vast majority" of the information is available only on computer tapes, according to Michael Garland, director of user services for the bureau.

Garland said the first of the detailed state-by-state tapes will be released in the next few weeks. The bureau expects all that information to be out by fall.

In 1972, the bureau sold computer tapes of its work product for the first time in all its detail. Several organizations were created to work the mother lode of demographics, including Claritas Corp. and the Marketing Analysis Division of CACI Inc., both in Rosslyn, Va. he intervening decade, and the attendant economic malaise, has made the information even more important, according to firms in the industry. While the economy was robust, marketers could afford to go on hunches, said Jonathan Robbin, chairman of Claritas. Now, "instead of sitting back and watching the pie get bigger, you have to fight for a bigger share of the same size pie."

Direct mailers and credit card companies want to know where their best customers live. Retail chains want to know where to concentrate their sales forces. Services like Arbitron and A.C. Nielsen Inc. want to know what size audience to sample in order to rate radio and television stations.

Claritas has specialized in matching zip-code geography with census and other demographic data to show advertisers how to reach their best customers. Robbin said the theory behind Claritas is that "people are tribal, territorial, and socially hierarchical." As a result, a zip code becomes a reasonable proxy for the type of consumer in a neighborhood.

CACI offers customers an on-line computer system that spits out the demography and buying power surrounding any particular street corner in the United States. CACI has also developed a buying-power mapping service.

CACI vice president for marketing services George Moore said CACI spends about $1 million a year to keep the information up to date. He said CACI's estimates were within 9 percent either way of the latest census numbers.

Both Claritas and CACI joined NPDC and 13 other companies in a consortium to purchase from the census all the demographic details by zip code. The group is paying $250,000 for the project, which will be delivered along with the income tapes this summer.

In Washington, there are two newcomers to the field whose founders came from the Commerce Department: CEC Associates, started last year by Courtenay Slater, former chief economist at Commerce, and Cenex Inc., established by four former Census Bureau employes, including former deputy director Daniel Levine.

The firms depend on the census to be right, but it does not always hit the mark. When the census had finished counting the population this time around, it found about 5 million more people than it had expected. Uncle Sam was 2 percent off overall, and as much as 25 percent off in some places, said Bryant Robey, editor of American Demographics magazine.

That is a problem for pollsters, according to former census deputy director Levine. When pollsters plan a representative sample, they have to know what the sample is supposed to represent. The unanticipated growth in population was mostly among blacks and Hispanos, according to Levine. Polls that spell out attitudes by race could be off as a result, he said.