With the national census still five years away, the government is already preparing to test new forms and methods aimed at getting a more accurate total than ever before.

Attempting to reach every American has become such a complex job that planning starts far in advance on how to find people, what to ask them, how to design the forms and how to learn the most while imposing as little as possible.For the most part, people like to be counted, and they like to read about the results, says Martha Farnsworth Riche, the new director of the Census Bureau.

"Census data is the stuff of people's lives," said Riche, a former editor at American Demographics, a magazine focusing on population and economic statistics. "People are fascinated by how they stand" in comparison to their neighbors and other communities.

But counting everyone is a struggle, she said. Forms are designed more for the convenience of data-collecting machines than the people answering the questions.

The bureau will begin testing new methods and forms in March, in Oakland, Calif., Paterson, N.J., and six parishes in northwestern Louisiana.

Someday, the census may be taken by computer. But the 2000 count will still be done mainly by mailing out forms and asking people to send them back, Riche said. Changes will focus more on improving forms and following up to count people who are missed on the first pass.

In 1990, workers visited some homes six or seven times before finally resorting to information from neighbors or others about people they couldn't reach - a costly and not especially accurate process.

So, one thing to be tested is the idea of doing a large survey of areas being counted. Then, if one or two visits to homes that didn't answer fail to get needed information, the survey could be used to estimate the typical person or family there. That test program is called "sampling for non-response follow-up."

Another problem is the undercount, or the number of people missed by the census, which statisticians say is a problem that tends to affect minorities disproportionately.

"We have always had an undercount, we always will have one," because people refuse to cooperate for various reasons, Riche said.

"But we can't count them out, because they are here."

This will involve another special survey, called "integrated coverage measurement," that statisticians will use to estimate the number and characteristics of people missed in the census. This estimate could then be included in the final count to provide a single more accurate number at the end of the census.

Such estimates were developed in 1990, but under political pressure, the bureau released two sets of numbers, the actual count and an adjusted number including the estimate of those missed.

That led to lawsuits and bickering between states and cities because using one set or the other could shift seats in the House of Representatives and millions of dollars in federal money from one area to another.