George Bush was supposed to be trounced in Iowa in 1980. Ten days before the caucuses, a Des Moines Register poll rated him third, well behind former California governor Ronald Reagan and then-Senate Minority Leader Howard H. Baker Jr.

But Bush won, squeezing past Reagan by 2 percentage points in the final tally and beating Baker by 22 points, margins that gave underdog Bush enough momentum that year to knock Baker out of the race and eventually do so well that Reagan asked him to join his ticket.

Predicting the outcome in Monday's Iowa caucuses is "terribly tricky," in the words of Democratic pollster Peter Hart, who is not affiliated with a candidate this year but polled for former vice president Walter F. Mondale in 1984. "Picking the Super Bowl looks like a piece of cake next to picking the odds in Iowa," Hart said. But the trickiness of the task has not deterred would-be prognosticators; pollsters have polled Iowa more intensely than ever this year, and the predictions are flying.

Iowa is complicated for several reasons: The rules governing the Democratic caucuses are complex, the number of people who will participate is tiny, Iowans are proudly defiant of national habits and trends, and they have a reputation for making up their minds at the last minute.

"Iowa is the ultimate sand trap for pollsters," said Republican pollster Richard Wirthlin, who worked for Reagan in 1980 and is advising Senate Minority Leader Robert J. Dole (R-Kan.) this year. "Anyone who thinks they know who's going to win by what margin simply doesn't know."

"We aren't measuring active voters and how they feel about the two candidates," he said, referring to Vice President Bush and Dole. "We're measuring active voters who will get up and go to a caucus maybe in the middle of a snowstorm, and there's no measure really to nail down precisely who those individuals are."

Of 1,200 voters interviewed in a January Gallup poll, fewer than 100 Democrats and 100 Republicans said there was a "high" likelihood that they will attend a caucus.

Samuel Popkin, a political scientist at the University of San Diego, estimated that about half of Iowa caucus voters do not make up their minds until the last two weeks. And "once you walk in the door {on caucus night} you might change your mind again," he said. "There are real contagions that can go on."

But Popkin said he thinks that there are early signals that strategists can use to determine who is likely to surge in the final days, especially as voters start to look for a candidate who seems "presidential" and likely to win.

Examining such factors would have provided a tipoff to Bush's late surge in 1980 and to Gary Hart's similar 1984 Democratic surprise. Hart doubled the percentage of the vote that the final polls said he would receive, ending Ohio Sen. John Glenn's campaign and achieving an unexpected, if distant, second-place finish to Mondale.

In 1980, Bush benefited from a strong field organization and from a good performance in a televised debate held shortly before caucus night that Reagan boycotted. This year, even Bush's competitors concede that his Iowa field organization is among the strongest in the state.

It also seems to help if a candidate can leave a good impression with voters whose opinions are fluid until the last minute. Hart, for example, was viewed the most favorably of all the Democrats in 1984. He had an 11 to 1 favorable-unfavorable ratio in the same poll that projected -- incorrectly -- that he would be in a close race for second place with Glenn and Sen. Alan Cranston (D-Calif.).

Poll results also have shown that a candidate can be crippled by a high unfavorable rating, as happened with Glenn in 1984 after a poor Democratic debate performance.

This time around, Hart is saddled with the highest unfavorable ratings of all the candidates, according to the KRC Communications poll conducted recently for Boston's WBZ-TV and Des Moines' WHO-TV. Rep. Richard A. Gephardt (D-Mo.) enjoys the best favorable-unfavorable ratio this time and has other potential strengths, including the high marks voters give him on agricultural issues.

Such secondary factors may give Gephardt a slight edge in the Iowa Democratic race, in which he, Sen. Paul Simon (Ill.) and Massachusetts Gov. Michael S. Dukakis are bunched at the top, according to most recent public and private polls.

In a fluid situation where voters often change their minds after being lobbied by friends and neighbors on caucus evening, pollsters have found that it is also important to determine who a voter's second choice is likely to be.

Front-runner Gephardt fares well by that measure, according to the KRC-WBZ poll, which shows that less than a week before the caucus vote, Gephardt was cited as the favored second choice of supporters of both former Arizona governor Bruce Babbitt and Dukakis.

Simon, who rated third in voter preference little more than a week ago, apparently has benefited from last Sunday's Des Moines Register endorsement and a last-minute spate of broadcast advertising.

Gephardt also seems to benefit from the good feelings that Iowans have for him. Forty-nine percent of those polled said that Gephardt "understands Iowa." Only 4 percent said the same for Dukakis.

Political analysts searching for a way to measure voting preferences in Iowa are constantly refining their methods. Lance Tarrance, pollster for Rep. Jack Kemp (R-N.Y.), for example, may use four different threshold questions to weed out those voters less likely to attend the caucuses.

Even then, he acknowledged, uncontrollable factors such as the weather and the composition of the caucus night crowd can affect the result.

For every candidate, there is a best-case scenario for caucus night. But the politicians working the bars, the telephones and the campaign stump in Iowa this week are hedging their bets and working to lower expectations there, just in case.

"It's like having a broken altimeter on an airplane," said one. "We don't know where we are. No one knows. We're using as good a technique, as good a methodology as there is available, but the altimeter is still broken."

Staff researcher Colette T. Rhoney contributed to this report.