For one very good reason, I believe the majority of electronic voting machines are fundamentally flawed - it is hard to give a reasonable guarantee that:
electronic records are written once and from that point read-only.
This is not an impossible problem, for example, modern database management systems can enforce such permissions (assuming there are no bugs affecting this functionality in the DBMS). Ideally this kind of functionality should be implemented in hardware, but that is generally more expensive.
In short, making such systems secure and reliable is a fundamentally hard problem.
Perhaps the best example of a voting machine done (almost) right, the Sequoia AVC Advantage, is well explained on Security Now! Episode 211 - Hacking Electronic Voting Machines:
Main Page - http://twit.tv/sn211
Transcripts - http://www.grc.com/securitynow.htm#211
The best documentary I've found explaining the problems with Diebold and many other machines is the HBO Hacking Democracy Special:
Google Video - http://video.google.com/videoplay?docid=7926958774822130737#
Main Site - http://www.hackingdemocracy.com/
The Princeton University findings are also good:
http://www.youtube.com/watch?v=aZws98jw67g
And the reason I remembered to write this post, a great post linked by the Electronic Frontier Foundation on Twitter:
http://www.freedom-to-tinker.com/blog/jhalderm/hacking-dc-internet-voting-pilot
Fundamentally in electronic mediums, things like trust are still being developed and are often at best fragile and relatively easily broken and overridden. That's the reason the heart of democracy still in my opinion requires a paper trail, literally a reliable, stable audit trail of every single vote - because paper is still a known medium that is much easier to protect and much harder to tamper with.
