Very interesting discussion. I can describe my set-up as a university professor, but its mainly a big mess, so its a do-not-do-like-me warning more than anything.

My scientific papers are mostly on paper, because when I started grad school in 1994, less than 50% of interesting articles were online in PDF format or my poor university did not subscribe to the online version. And if they were, I took the habit of printing them anyway to read on the bus, at my desk, outside on the grass, etc. I stayed with paper out of habit. I started my master in 1994, added a PhD over that and a 4 year post doc, so now I have about 20 legal size boxes of reference material I cannot bring in my new office. They are sitting in my basement since I started as a faculty back in June 06, and I still haven't figured out what to do with them. The irony is that I kept to paper for now, which makes my problem bigger and bigger, but at least I resist the urge to print much more since I apply GTD, because if I print, I decided I must assign it to a next action or file it in a project reference material box (the kind of cutted halfway boxes to hold magazines you get at IKEA) which get reviewed weekly (they are like super inboxes). I presently have 8 boxes like these on my desk, labelled with my main current research projects or class I teach. These boxes MUST be part of the weekly review, because if not, I would end up dumping papers in them and not doing anything about them (reading and note taking) and not moving the project forward.

My seminar notes are in small Clairefontaine books (aaaah silky smooth writing) that also act as brain dumps. I keep them on a shelve but they are very very hard to sift for interesting information or project idea I had while I was not listening to a speaker or riding the metro or waiting in line at the post office. I moved to 3*5 cards when implementing GTD last summer, so the 1-card-for1-talk is much easier to file as general reference material in one of my 8 desk inboxes or the general inbox for later processing during the weekly review.

I am very interested by Devonthink after seeing the Steven Johnson example that Berko gave, because the 8 boxes I have definitely overlap in topics (a paper on aggressive behaviour or hormone X or changes in gene expression with Y variable could be interesting for 3 of them), so all this clunky filing may change, except that I will still have the equivalent of 5 mature trees in my basement in the form of research papers on subjects I do not work on anymore. I like geeklady's advice of having broad topics (if you go for paper notes) because some things do not seem to have an interest for a given part of your thesis up until you come across it and read it again 2 months later (or read teh note syou took on it) and see how relevant it really is for your fourth chapter.

So the short answer is: don't do like me and take the good advices of the smart people here, which you are obvioulsy already doing since you started this thread.

I am going to research devonthink right now