Just before the Pennsylvania Primary, the race was seen, by most pollsters, as "tightened". Generally, they showed around a 52% to 48% advantage; Clinton over Obama. On the day of the primary, the exit polls showed a near similar tightened spread. Then, when the votes were counted, the polls were clearly shown to be wrong. Instead, it is as if the margin of error was added to Hillary's advantage and equally subtracted from Obama poll numbers in order to get the right answer. That's how far off things were.
The pollsters have definitely got a problem. At best, they just think that Obama supporters are more enthusiastic and, therefore, want to express themselves more freely to the pollsters. Others say that there is a dark side, a more sinister problem, that involves whites not wanting to admit that they didn't or won't vote for Obama. The fact is, some whites might feel "subliminally" racist if they "say" they didn't vote for the black man. For that reason, they either avoid the pollsters or simply lie to them.
Whatever is happening, it is obvious that the polls aren't accurate. Right now, Obama is seen to be leading McCain in most of the head-to-head national polls. Can we really trust that? If it is anything like what was happening in Pennsylvania, Obama is probably "overstated" by the margin of error and McCain equally "understated". If so, those polls are probably really saying that McCain is ahead by a two to three percentage points.