I had a little trouble with this too. JMDict is the new format, so you should go with that. I think they were or perhaps still are updating EDICT2 in unison, but you can basically think of it as deprecated. There's a DTD for JMDict and a page explaining the relational data model they use themselves for JMdict somewhere on the site.
It took me quite a bit of work to parse it successfully into a DB, but then I was using a fairly low-level Java library. There are a lot of elements, so I'd study the DTD and the data model page before writing any code. Since JMdict is a relatively new format, the Bibliography, Link and Audit tags are currently completely empty, so I wouldn't prioritise those for now.
If you're making a J-E dictionary, you'll probably be interested in Kanjidic2, the Tanaka Corpus, Kradile-u (check license), KanjiVG, and some kind of radical list (I had to make my own).
It took me quite a bit of work to parse it successfully into a DB, but then I was using a fairly low-level Java library. There are a lot of elements, so I'd study the DTD and the data model page before writing any code. Since JMdict is a relatively new format, the Bibliography, Link and Audit tags are currently completely empty, so I wouldn't prioritise those for now.
If you're making a J-E dictionary, you'll probably be interested in Kanjidic2, the Tanaka Corpus, Kradile-u (check license), KanjiVG, and some kind of radical list (I had to make my own).