This National Review piece makes a common argument that there could be lots of voter impersonation fraud but it is just too hard to detect.
I take this argument on directly in my forthcoming book, The Voting Wars. Briefly, the argument is not persuasive for two reasons. First, impersonation fraud is an illogical and inefficient way to steal elections (compared to, for example, vote buying through absentee ballots). Second, other types of fraud (such as vote buying through absentee ballots) are in fact caught and prosecuted, and there is no reason to suspect that impersonation fraud would be harder to catch. Indeed, it should be easier to catch, because it would involve a vast conspiracy of people showing up at the polls claiming to be either someone on the rolls or a falsely registered person (who under HAVA would have to produce ID as a first time voter).