As they wandered between discussions of why people don't answer their questions or how the Internet will affect their business, the nation's leading pollsters continually confronted evidence of their worst failure: the presidential election of 1948.

The American Association for Public Opinion Research was only a couple of years old when the polling business suffered the Truman embarrassment. The field's dominant figures, Archibald Crossley, George Gallup and Elmo Roper, declared as a certainty that Thomas Dewey would defeat Harry Truman. They were as sure of it as their descendants (and most leading Democrats) were in 1991 that George Bush was unbeatable.So when the association gathered here for its 53rd annual convention to discuss current problems and anticipate future issues, the pollsters faced a large exhibition on the 50th anniversary of the debacle. The exhibition served, as James Beniger, the group's president, said, "as a moral lesson for public opinion pollsters: Be humble."

In 1948, pollsters had a lot to be humble about, having foreseen Dewey as the winner by 5 to 15 percentage points. Truman won by 4.4 percentage points.

By and large, they faced up to it, the exhibition showed. There was Roper saying his face was "just as red as President Truman said it would be." In his whistle-stop campaign, Truman had scorned the polls as much as he had scorned Dewey, then the governor of New York. "We were wrong," Gallup said. "Bitter as the experience was, we intend to profit from it."

And profit they did. In fact, commercial success sustained the polling industry, because American business was already reliant on market research. Some commercial clients dropped off for a time, but the impact was short-lived. Dwight Eisenhower's 1952 landslide, anticipated by all the pollsters, sealed the restoration of their image.

Only a handful of 200 or so participants in this year's meeting had any connection to the polling of 1948. But one of them, Helen Crossley, remembered talking long-distance with her father from Denver, where she was a graduate student. She said Archibald Crossley was worried that Truman might be making a comeback, but told her, "You have to go with your data, not your instincts."

Irving Crespi, a college teacher in 1948 but afterward a longtime Gallup researcher, said, "The real villain of 1948 was Paul Lazarsfeld," a leading scholar whose research in Sandusky, Ohio, in 1936, 1940 and 1944 demonstrated "that elections are decided by Labor Day."

Crespi said Lazarsfeld "advised Roper not to bother with the last poll." He said he had been told that the Gallup organization had planned to poll in the last two weeks, but "Gallup talked to Roper and decided `why waste the money?' "

The exhibition demonstrated how the pollsters were not alone in their certainty. It showed the cover of the Nov. 1, 1948, issue of Life magazine, featuring Gov. Dewey, with the caption: "The next President travels by ferry boat over the broad waters of San Francisco Bay." It also carried the front-page banner headline from the Nov. 2 issue of The Washington Post: "Dewey Deemed Sure Winner Today."

There were documents showing how the pollsters changed their practices after the 1948 mistake. The most obvious changes were to state their findings with a little less Olympian certainty, and to continue polling until the weekend before the election. (Now, some pollsters, shaken by how Ronald Reagan turned a close race into a landslide in the last days of the 1980 election, keep going through election eve.)

But the most profound change was in how the pollsters selected the people they questioned. In 1948, they used quota sampling, where interviewers were told to find so many young white men, so many elderly black women, and so on. The interviewers then picked the individuals, sometimes on street corners.

For some time, pollsters had been urged by academics to shift to probability sampling, which sends interviewers to homes chosen by chance around the nation, and at least theoretically gave every American a chance to be polled. After 1948 pollsters took the professors' advice, and switched.

For some of today's pollsters looking at the old speeches and reports, the exhibition was an intriguing introduction to a history they knew only vaguely.

But to Kathleen Frankovic, director of surveys for CBS News and a past president of the polling group, the exhibition contained a very modern message, one that transcends the methods of 1948 and the changes that followed.

"In 1948, the error was overconfidence and belief in oneself" and the perfection of current methods, she said.