Rasmussen Reports (6/16, likely voters, no trendlines):
Arlen Specter (D-inc): 51
Joe Sestak (D): 32
Other: 4
Undecided: 13
(MoE: ±5%)
This is the first time Rasmussen is dipping its toe in the water here, so we don't have any kind of trendline to work with. But Specter's 19-point lead is the smallest any pollster has shown to date. (A GQR survey put Specter up 55-34). Given how far off Pennsylvania's primary is, that doesn't strike me as a terribly formidable margin, especially since Specter is so much better-known.
Rasmussen's favorability numbers are a bit surprising, though. Among Dems, Specter clocks in with a 72-26 rating, not too different from the six-week-old R2K poll taken for Daily Kos. However, Sestak's 57-21 favorables seem way high. By comparison, that same R2K survey (which was also of likely voters) showed 56% having no opinion of the guy, as opposed to just 22% here. A more recent Quinnipiac poll (of registered voters) showed even bigger d/ks, as did a Republican survey of LVs.
My guess is that this difference comes down to methodology. All prior polls taken of this race used live interviewers; Rasmussen uses IVR, or "interactive voice response," aka robopolling. Obviously the discrepancy is because the DOG COULD HAVE BEEN ANSWERING THE CALL. I keed, I keed. Really, it's just that lower undecideds across the board, whether for favorables or head-to-heads, are simply a hallmark of the push-button nature of robopolls.