Under normal conditions, statistical sampling is highly reliable. Done correctly, its accuracy can be measured to within a range of a few percentage points. But when politics enters the picture, the conditions are hardly normal.

Americans have had enough of politics robbing fairness. A government that can't pass meaningful campaign finance reform should not rely on anything less than a head count in the 2000 Census.That kind of logic doesn't sit well in Washington, where some have re-sorted to calling the need for statistical sampling the biggest civil rights question of the age. To them we say, take a deep breath, holster your race cards, and get a grip. The only unquestionably accurate way to count people, especially when congressional districts are at stake, is one at a time.

Of course, the president and most Democratic lawmakers say the one-at-a-time method isn't fair to minorities and inner-city Americans who are too difficult to reach. By some estimates, the 1990 census left out 6 million of these people. Predictably, this has become a partisan political battle.

Most of the Democratic arguments have lost power in the face of the conventional wisdom that a majority of the uncounted Americans are Democrats whose presence would help that party gain more seats in the House. Under-count Republicans in 2000 and guess which side argues for sampling next time.

Politics aside, no one can argue credibly that statistical sampling would do anything to end the fight over accuracy. If anything, it would add more uncertainty. The only answer, simplistic though it sounds, is for the census to do whatever it takes to count every single American.

That's what the House voted to do last week, although the fight is far from over. Coincidentally, this week's vote took place right after a census trial run on a Wisconsin Indian reservation - another of the many areas in the nation where people go undercounted. Initially, the bureau handed out questionnaires, but only 41 percent of the people returned them. Later, however, the bureau visited each home that failed to return the documents and ended up accounting for each household.

The exercise demonstrated two things. First, census taking is not easy. It doubtless was hard in 1790, as well, when the nation was mostly rural. Despite an energetic education campaign, most people disregarded the forms. Second, workers can achieve accurate counts if they are persistent.

Already the census bureau has wisely decided to make the 2000 forms simpler and less intrusive. Why, for instance, did the old forms have to ask questions about income and the market value of a home? These are now missing, which will no doubt help workers deal with a populace that seems to be increasingly skeptical of the federal government.

But, in the end, census workers must deal with the people. The politicians have proven they can't act or think straight when their own money or power is at stake, and they should butt out.