WASHINGTON (AP) The Census Bureau refused to release statistically adjusted census data to disburse billions in federal dollars, foreshadowing a possible legal battle with Democrats and big-city mayors who say raw figures undercount minorities, the poor and children.

Explaining the bureau"s decision yesterday, Acting Director William Barron said agency researchers discovered errors in census data that had been adjusted using sampling methods.

That leaves the raw figures from last year"s national head count as the official data on which the government will base decisions on the distribution of more than $185 billion to states and local communities for Medicaid, foster care and other social service programs.

Barron said the errors were so fundamental they prevented the use of the sampled results "in their current form."

A recommendation to Barron from a committee of career bureau officials was the basis of his final decision. Barron said more research would be needed before determining if there will be any public release of the adjusted data but did not say when that decision would be made.

Democrats immediately criticized the agency.

"This is an abysmal decision from a bureau whose mission is to count people accurately in this country," said Atlanta Mayor Bill Campbell, who said he would consider suing to get adjusted figures released.

The bureau faced a similar decision in March, when it recommended against adjusted data as the basis for redrawing congressional, state and local political districts.

There were too many discrepancies among adjusted data, the actual count and a third survey used to measure accuracy, and not enough time for further analysis, Barron said at the time.

Rep. Dan Miller (R-Fla.), chairman of the House census subcommittee and a vigorous opponent of sampling, supported Barron.

"As we move forward toward the 2010 census, we must not repeat the mistakes of the past," he said. "It"s time to put adjustment, for political purposes, to rest."

Several lawsuits already have been filed over the conduct of the census. One suit, filed by Oregon state senators, said adjusted data could account for about 43,000 people missed statewide.

Another dozen or so cities and counties have joined a lawsuit spearheaded by Los Angeles over the Census Bureau"s initial decision to use the raw head count for redistricting.

Jessica Heinz, assistant city attorney for Los Angeles, said her office would consider another suit over yesterday"s announcement. Los Angeles Mayor Jim Hahn is a Democrat.

Democratic mayors have been most vocal about getting adjusted data released, although many large cities currently are headed by Republicans.

The use of adjusted numbers has long been the topic of partisan debate and legal battles.

The administration of former President Bush faced a similar predicament 10 years ago when then-Commerce Secretary Robert Mosbacher overruled the Census Bureau"s recommendation to use sampled data. Though acknowledging minorities were undercounted, Mosbacher rejected the second set of numbers on the ground that it would "abandon a 200-year tradition of how we actually count people."

Cities including New York, Los Angeles and Detroit sued over the 1990 decision and the adjusted numbers. All the adjusted 1990 figures were released eight years later.

Generally, Republican opponents claim sampling would insert more errors into a 2000 census that already was better than the one in 1990, because of a lower national net undercount. They have also said that while adjustment may count people originally missed, it may not place them in the correct neighborhoods.

The bureau had identified in March a net national undercount in 2000 of 1.2 percent of the country"s 281 million people, or about 3.2 million. The 1990 undercount was 1.6 percent, or about 4 million then.

Census officials have conducted more research since March, Barron said yesterday. As a result, he said, preliminary estimates indicate the net undercount could have been reduced to as low as .06 percent _ by far the lowest undercount in history.

Specifically, yesterday"s decision determined whether adjusted population counts would be used for purposes other than drawing new political boundaries. It affects 2000 census data yet to be released, as well as various other population estimates and surveys the Census Bureau conducts between once-a-decade head counts.

The results from many of those estimates and surveys are used in formulas that determine distribution of federal money to states and local communities.