As a poll junkie, I generally consider Survey USA and Rasmussen to be among the most reliable (with Zogby having about the accuracy as making something up). On their site, you can check their election track record and its pretty good.
In terms of average error on the margin of victory in 2004, SurveyUSA was second only to Mason-Dixon. They had the 2nd-best percentage of picking the correct winner, behind only Rasmussen (and beating Mason-Dixon). None of SurveyUSA's polls were outside the margin of error in 2004, but 3 percent of Mason-Dixon's were and 5 percent of Rasmussen's were.
Anyone who says that SurveyUSA is not reliable is simply uninformed and/or unwilling to face reality.
Also I would add that Survey USA was the best pollster in 2004 in terms of partisan bias. Their average bias was .1 percent in the Democratic direction. That is the closest to even among all pollsters that conducted more 5 statewide polls in 2004. (For comparison Strategic Vision was biased toward Republicans by 1.9 percent, Mason-Dixon was biased toward Democrats by one-half of a percent, Rasmussen was biased to Dems by 2.2 percent and Zogby was biased towards Democrats by 3 percent)