Even in a shaky market, there may be a fairly safe way to make money: Look at the polls on the Democratic primaries for the American presidential election and bet the other way.

It has worked remarkably well during the last five weeks. An online trader who placed a $100 bet on Hillary Clinton on the day of the New Hampshire primary got a payoff as high as $10,000 after the polls were proved wrong and she won a surprise victory.

Likewise, anyone who bet on Barack Obama winning a much larger-than-expected victory in South Carolina would have done very well. And last Tuesday, a bettor with access to the notoriously unreliable Super Tuesday exit polls, widely leaked before the voting was complete, could have given points and gotten rich. This is a commentary on the state of polling in America today. Even more, it reflects a dynamic Democratic Party electorate.

“It always is difficult to poll primary voters, particularly in an environment like this with a huge turnout among women, blacks and young voters,” president of American Viewpoint in Alexandria, Va., and a longtime Republican poll taker, Linda DiVall said. “All the models are totally skewed.”

Iowa, site of the first caucuses, is illustrative. One of the few pollsters to get it right was Ann Selzer for the Des Moines Register, who forecast a big Obama victory. This was predicated on a huge turnout, something the other polls and most every politician in the state dismissed.

On January 3, 240,000 Democrats came out on a cold night to caucus, almost double the turnout of four years before, and, as Ms. Selzer forecast, Mr. Obama won a decisive victory.

The pollsters botched it again five days later when all predicted a big Obama wave in New Hampshire. This time it was an outpouring of women for Mrs. Clinton that produced an upset and lots of red-faced pollsters.

It hasn’t gotten better. Last Monday, John Zogby, in a poll for C-Span, the public service cable-television network, and Reuters, projected that Mr. Obama would carry California by 13 points; Mrs. Clinton won it by 10 the next day.

SurveyUSA had Mrs. Clinton winning Missouri by 11 points; Mr. Obama carried it by a point. The same survey projected an even race in Alabama, where the Illinois senator won by almost 15 points.

Much of any explanation has to do with the unique American primary system: Americans are generally less moored to political parties than are Europeans, and primary voters are especially unpredictable. Most British or French polls focus on general elections, and the outcome is usually determined by a small minority of swing voters. In American primaries, often large numbers of voters swing back and forth between candidates.

There are, to be sure, problems with polling. New Hampshire occurred only five days after Iowa, and the surveying for the second contest was halted too soon. With Mr. Obama, some experts speculate there may be some closet racism, with respondents unwilling to admit they wouldn’t vote for an African-American. Yet his showing in most states undercuts that case.

There are dubious methodologies. SurveyUSA does automatic robot calls, where you push a button to respond. “Your dog could answer the call,” Ms. DiVall said.

Every poll, good and bad, is featured daily on fabulous (for political junkies) Web sites such as RealClearPolitics.com and Pollster.com. Averages include all of them, sometimes producing a Gresham’s Law of polling, with the bad ones driving out the good.

There’s also the issue of cell phones. By this Election Day, more than 20% of American voters may not have landline phones, yet the standard telephone-dialing systems used by most pollsters include very few cell phones.

Most reliable studies indicate there’s little difference between cell phone and landline users on political and ideological questions. Yet a big boost in turnout of young voters, who disproportionately use cell phones, is making the best pollsters a little nervous.

Susan Pinkus, who conducts the Bloomberg/Los Angeles Times poll, says she’s exploring a “multi-modal” survey that combines landlines and cell phones. As a result of all the polling controversies, the American Association for Public Opinion Research has commissioned a study to see what went wrong.

That may be useful. A much more important story than flawed polls or methodologies, however, is the tremendous turnout in Democratic primaries. This makes obsolete most of the pollsters’ models — in which previous voting behavior is considered a predictive measure of likely voters. As Ms. Selzer realized in Iowa, this contest is generating a new universe of voters.

This is true in “blue” (Democratic) and “red” (Republican) states alike. In reliably Republican South Carolina and Georgia, the Clinton-Obama turnouts were almost double the presidential primaries of four years earlier and larger than the Republican vote. In Alaska, one Democratic caucus venue was shut down by a fire marshal; there were too many people there.

This is because of an unprecedented influx of young voters, African-Americans, and women, the data suggests. It’s being generated by both camps, though most of the energy resides with Mr. Obama. Last Tuesday, 200,000 more Democratic voters turned out in Illinois than in New York, which has a much larger population.

In 2000, the polls were wrong in New Hampshire, where a similar outpouring of voters gave John McCain a big win over George W. Bush. This year, the surveys have gotten most of the Republican primaries right, or within the margin of error. That’s good news for the pollsters if not for the relatively unenthusiastic Republican constituencies.

Several weeks ago, a Wall Street Journal headline declared: “In Campaign 2008, Pollsters Are Biggest Losers.” Try telling that to Rudy Giuliani or Fred Thompson.

Mr. Hunt is the executive editor for Washington at Bloomberg News.