The Bradley Effect is the name given to the idea that a significant number of racist voters will vote against a black candidate but be unwilling to admit that to pollsters. A narrower version of the conjecture is that they will tell a pollster they are "undecided" even though have fully decided to vote against the black guy. The eponymous effect is attributed to the fact that Tom Bradley lost the California governor's race to George Deukmejian in 1982 even though some late polls projected Bradley the winner, and the prestigious Field Poll projected a 7-point margin for Bradley.
V. Lance Tarrance, Jr., who was Deukmejian's pollster, debunks the conjecture here. (Thanks to John for the link.) Except for the Field Poll, the polls showed that the 1982 race was tight—within the margin of error. Tarrance's polling showed the Bradley margin steadily declining in the last five weeks, with Bradley ahead by a statistically insignificant 45-44 in the last pre-election poll. Other experts looking at exit polls and results in select precincts projected that Deukmejian had won. Mervin Field said after the 1982 election that "race was a factor in the Bradley loss." Thus, "Field Effect" could be an eponymous term for a humiliated researcher blaming his data.
Polling analyst, Nate Silver, points out here and here that it is impossible to measure a lying-to-the-pollster effect, but points out that there have been recent races involving an African-American candidate in which there was no significant discrepancy between late polls and outcomes.
Skeptic
Bradley's campaign manager in 1982, Nelson Rising, says here that their internal polling showed Bradley's lead shrinking as election day approached, and he denies there was any "Bradley Effect." Another Bradley campaign insider writes here that there was no "Bradley Effect." Thanks to John for these links too.