I recently discovered Mendeley, a reference database / science social network site. There are several cool features (and a few misfeatures). First the desktop software is backed by a website, so you can synchronize your article metadata across computers. For me, this is essential and why I do not use desktop reference managers. (I have about 1,200 articles listed in a bibtex file which I synchronize between my laptop and the lab file server.) The other cool feature is that it runs across platforms, another essential feature in my book. Perhaps the best feature of the software is that it can directly extract bibliography information from PDF articles, so no more typing them in! Also, you can store up to 500MB of metadata + articles on their site (and this is synch'ed across computers).
There are some strange features too -- when I loaded all 1,200 of my bibtex entries, many were not correctly handled. The problem seems to be with edited volumes as Mendeley expects an author list and not an editors list. (It stores the editor list, but displays the records as authorless which gets confusing.) The other wierd feature is when you go to their web page, it lists most read papers in different fields, etc. You cannot however click on the article titles and be redirected to a web lookup page. Kind of pointless to see what folks are reading, but have no way to click through to the article itself.
I'm optimistic that this will be useful software. However, I find that I rarely search my reference database. I only keep them for the purpose of inserting citations into my papers. Instead, I almost always use ISI and Google Scholar to find articles. On the other hand, I do tend to have PDF's pile up on my desktop and Mendeley would be a good way to organize them.