The U.S. Census Bureau has issued revised per-capita-income data for about 400 of the nation's 3,100 counties after finding that some of its clerks made errors when they took information from the forms in the 1980 census.

The corrections, which indicate that the data first reported for some localities was too high, will cause changes in revenue sharing and other federal grant programs where the money distributed is tied closely to census information.

Correcting the errors also is expected to hold up new reports by three or four months, officials said, in a census that cost more than $1 billion and already is about a year slower in issuing much of its data than the count conducted in 1970.

Although the mistakes have little effect on national statistics or those for large counties and states, their impact on some smaller places is substantial. For example, the bureau has cut $1,413 from the $6,574 per capita income reported for Appomatox, Va., and $1,065 from the $7,978 for Petersburg.

"When you have large numbers of people doing repetitive clerical work, something's bound to happen," said Peter Bounpane, assistant director of the Census Bureau for demographic censuses. "We're correcting all those things that are wrong. But next time in 1990 , we're trying to make the census far more automated than this one."

All the mistakes were in one direction, Bounpane said, making reported incomes too high.

In the Washington area, according to the census correction notices, per capita income data issued last summer was overstated by $52 for Fairfax County, $32 for Alexandria, $9 for Montgomery, $3 for the District of Columbia, and $2 for Arlington.

Figures for all other counties and cities in the area are unchanged, but census officials said corrections will be issued within a few months for neighborhoods and other small areas, called census tracts, in jurisdictions where overall changes have been made. Until then, they said, it is uncertain which income data for small areas is correct and users should be cautious.

Bounpane said the mistakes occurred as clerks pored through the long forms, containing detailed information, filled out by about 15 million of the nation's 80 million households.

On the questions asking income in 1979, he said, clerks were supposed to round off amounts to the nearest $10 and take off the last zero before filling in circles that could be read by optical scanners and fed into computers.

"Unfortunately, sometimes they kept in those last zeros," Bounpane said, "or they filled in the circles from the left so the amounts are too large by factors of 10 or 100 or 1,000."

Bounpane said the Census Bureau believes the mistakes affected only about 0.5 percent of the forms nationwide. But where they occurred is very uneven, depending on what particular clerks did. The greatest number of mistakes, he said, seems to be in the South.

In the worst case found so far, 8.6 percent of the forms were incorrectly coded in Treutlen County in rural Georgia. The Census Bureau has lowered its reported per capita income by $3,843. In Taylor County, Ga., the mistake came to $3,602 per capita.

"It hasn't made much difference for big places where you have enough good data to keep the errors from changing the results," said Paul Manka, an offical of the census data users service. "But if you put a millionaire in a small town, it can really distort things."

Bounpane said the Census Bureau first started looking at its income data again in early fall after complaints from about 200 local governments across the country that their revenue-sharing grants were too small because their reported income seemed too high. Since October, he said, workers have double-checked several hundred thousand microfilmed forms.

The first notice that there might be a problem with the income data was sent to purchasers of Census Bureau computer tapes in November. A few weeks ago the purchasers were sent a second notice, listing affected states and counties. A small item explaining the problem was printed in The Numbers News, a newsletter published by American Demographics Inc.