THE DECENNIAL CENSUS taken next April is bound to be wrong. It should be closer than the 1970 census, which missed perhaps 5.3 million people. But some undercounting is inevitable. And its rate is likely to be highest in cities and among low-income people, minorities and aliens - groups that tend to be most wary of official inquiries.

This would be bad enough if the census were used only for its original purpose, apportioning seats in the House. Now that census data are also used to allocate many billions in federal and state funds, an undercount means less aid, often for the areas and people who need it most.

The Census Bureau has elaborate plans for reducing undercounts by using community residents as census-takers, publicizing the importance of being counted, and emphasizing that people's responses will be kept confidential. Leaders in Baltimore and other cities are plannig to help.

Many mayors and urban groups want the bureau to go further and adjust the official 1980 returns to take uncounted people into account. Their pleas are expecially fervent because many cities are losing people anyway. According to a new Census Bureau report, the 30 House districts with the greatest population declines since 1970 are all urban, mostly Northern and Midwestern and largely minority or low-income areas. One Bronx district has lost 39 percent of its population since 1970. Rep. Parren Mitchell's Baltimore district has lost 23 percent. Reapportionment and redistricting are bound to diminish such areas' strength in the House and state legislatures while suburbs - and in Congress, the "Sun Belt" - gain.

Of course the urban losses will be larger, unfairly so, if their counts are badly incomplete. Demographers can no doubt calculate fairly accurately how many people have been missed. But is "correcting" the census a good idea? For federal apportionment purposes, no. That is a constitutional process. Article I calls the census an "enumeration"; the Fourteenth Amendment speaks of "counting the whole number of persons in each state." Those terms dictate use of an actual count - and great effort to make it as whole as possible.

Adjusting federal aid formulas is something else. For all their ring of equity and precision, those formulas are approximations anyway. For instance, the number of low-income children in a city is not a perfect gauge of its educational problems. Moreover, formulas are political instruments, continually modified and hedged t o allocate money for various purposes in ways that majorities of both houses find acceptable. Congress could certainly add in an "uncounted" factor, and in some cases probably should. That would be easier if one adjustment method were agreed on in advance. As Eddie N. Williams wrote in an op-ed article last week, a panel of the National Academy of Sciences has recommended this, but Commerce Secretary Juanita Kreps and the Census Bureau have been slow to respond. Mrs. Kreps should make a decision now, before communities start yelling about having been short-changed again.