For those of you who are still a bit fuzzy on the details of the electronic voting situation, here's a summary:
Electronic touch-screen voting was pushed as the answer to the "hanging chad" problem found with poorly-maintained punchcard voting systems. In the wake of the 2000 election, last October the US government passed the "Help Americans Vote Act" (HAVA), which gave financial support to voting districts to replace aging manual voting systems with modern electronic voting systems. While the current controversy rages, it's worth remembering that e-voting systems offered some important advantages over traditional methods:
The first two proved more-or-less accurate. The last two, not so much. Electronic voting systems are typically little more than stand-alone PCs running software on top of Windows. As we all know, software often has bugs, and Windows is not the most stable operating system around. Unsurprisingly, then, widespread problems with electronic voting and tabulation systems occurred in elections in 2002, 2003, and 2004. In some cases, polling units crashed, taking hours to fix (and functionally disenfranchising people who couldn't vote and couldn't return to the polling place later). In other cases, the systems offered grossly incorrect results, such as elections where the total number of votes exceeded the district's entire population. (A document detailing some of the more troubling examples of electronic voting system malfunctions can be found here (PDF).)
But software bugs leading to system crashes and obviously bad results are far from the worst possible problem. An examination (PDF) of an accidentally-exposed copy of the Diebold voting system source code revealed it was riddled with flaws which would make it possible for an unauthorized user to get into the voting system and make surreptitious changes. As all of the commonly-used voting systems are closed source, it's possible that the other big election software companies (ES&S and Sequoia) suffer from the same flaws. As they refuse to open up their code for inspection by software specialists, we have no way to know. Those of you knowledgable about computer security will recognize this as an attempt at "security through obscurity," and will know that it rarely works for long.
Making matters worse was DRE manufacturer misbehavior, which played a key role in convincing the California Secretary of State to de-certify Diebold machines. In a random audit of 17 districts using Diebold devices, none of them were using the version of the software certified by the state -- all had been updated or patched with code that the state had never seen or approved. While there was no evidence of vote tampering, there was also no way to know for sure.
The most commonly-offered solution to this dilemma are Voter-Verified Paper Ballots (VVPB), in which a paper ballot is printed when the vote is cast, inspected by the voter to make sure the paper reflects the actual vote, then deposited in a ballot box. The paper ballots can serve as either the backup (in case the electronic system malfunctions), as a comparison for spot checks of e-voting system accuracy, or even as the "real" vote, with the electronic tabulation used as the first estimate. In each case, the tally of the paper ballots would serve as the final arbiter of the election result.