The preliminary 1990 census numbers released last week could grow by 2 million to 5 million people before final figures are released late this year -- leaving in question until then the location of several congressional seats and millions of dollars in federal programs.

Although the Census Bureau has heard from or visited the vast majority of American households since it began its decennial head count this spring, there are still large segments of the population that have not been tabulated -- from military dependents overseas to big-city residents missed by census workers.

Although bureau officials will not speculate on how many people they believe remain uncounted, demographers, political consultants and congressional aides say the bureau's preliminary national count of about 245 million could reach 250 million. That would match the bureau's earlier estimates of the nation's population.

A significant number of those still "in the pipeline" are in households that were missed by the Census Bureau but who will be discovered in a process known as "post-census local review."

Local jurisdictions, which received preliminary housing totals from the bureau during recent weeks, can challenge those totals based on their own tax, utility or other records. The bureau has said that if it accepts the challenge, it will recount those housing units.

In New York City, for example, officials said yesterday that on each of 6,000 city blocks, they counted at least five housing units more than the number submitted by the bureau for those blocks. At stake are approximately 200,000 units citywide that could contain 500,000 residents and be worth more than $100 million in federal funding annually, said Hulbert James, the city's census coordinator.

If the bureau adds to the New York totals even half of the residents James thinks it missed, it could help the state keep one of the three congressional seats it has been predicted to lose in the 1990 reapportionment, said Mark Gersh of the National Committee for an Effective Congress, a Democratic group.

In addition to local review, the 1990 figures will also grow when the bureau completes a process known as "imputation." This is used when census workers have been unable either to contact occupants of a household or receive reliable information from neighbors. The bureau then uses a statistic average to "impute" the occupancy of that housing unit.

Bill O'Hare, a demographer at the Population Reference Bureau, said imputation added 760,000 residents in the 1980 census and could add as many as a million this time.

The preliminary numbers also do not include final figures for overseas military personnel and their dependents, nor the non-military federal employees stationed abroad. This could also add about a million persons to the population totals.

O'Hare said up to about 50,000 residents could be added by a bureau campaign designed to catch people who were missed altogether.

The preliminary figures are also missing any households not counted until mid-August or later. In some communities, census workers were still going door to door and keying their findings into the computer when the preliminary numbers were tabulated.

The figures could also rise as a result of internal reviews conducted by the bureau, in which the bureau checks addresses where census workers think they may have missed people.

Finally, population figures for a local jurisdiction could change when "split blocks" -- those blocks split between two jurisdictions -- are added. Residents of those blocks were included at state and county levels, but not at the local level.

Political consultants say that, because state populations could grow significantly before apportionment totals are released in December, five to 10 congressional seats could still be up for grabs.

These are in states -- including Montana, Kentucky, Massachusetts and New Jersey as well as New York -- that appear to have missed gaining or losing a seat by a relatively small margin. In some cases, population totals for these states fell well below the Census Bureau's own estimates, raising the possibility that many people were missed and that a seat could ride on the numbers still to come in.

"These statistics look surprisingly low," said Gersh. "The difference in the census estimates and the actual preliminary counts is very large, suggesting there is an undercount."