As we wait and hope that Japan can get the damaged reactors back under control, I'm already seeing a lot of chatter about 3 Mile Island and of course, Chernobyl.
Back in the 80s, in the wake of Chernobyl, CBC radio's program "Ideas" did a three part documentary called Counting the Costs on nuclear accidents. In 2009, they re-aired them (which is where I first heard them). This was a fanastic series that dramatically retells the stories of not just 3MI and Chernobyl, but a series of lesser known nuclear accidents through the 50s and 60s too. It really feels like a throwback to the era of actual radio programs like The Shadow and personally I found it enthralling. Particular to 3MI, the story includes not only the play-by-play of system failures and small oversights that lead to the crisis, but also includes the story of a federal nuclear regulator named Jim Creswell who spotted the very flaws that led to the accident ahead of time and was (of course) ignored.
You can start with any part, they're all good as standalone pieces, about 1 hour each.
Podcast part 1: Early accidents at the dawn of the nuclear power era
Part 2: Three Mile Island
Part 3: Chernobyl
My own take away from the 3MI piece was that the complexity of the systems needed to run a nuclear reactor safely make design flaws unavoidable, and thus accidents, inevitable. No great level of abject human incompetence is required. The usual sort of errors that even experienced and trained experts will make from time to time are enough. Well designed systems can survive a few of these errors, but not if they happen concurrently in ways that overwhelm the various safety margins built in.
Japan's situation is arguably different given the "act of God" aspect, but then earthquakes do happen and happen often enough that nuclear plants designed to run for decades can't claim to be shocked when struck. I'm hearing this is the 5th or 6th largest quake in "recorded history" - that's certainly big, but then we have records for even bigger quakes just in the decades we've been running seismographs. I will wait for more investigation and facts to emerge by my starting position is that whatever went wrong here is a design flaw.