I am still fighting with my huge library of pdf files. Since last post on this issue, I took a closer look to on-line reference managers. The biggest on the market are citeulike or connotea. The idea was to collect bibliographical data from the web instead of opening each of my pdf files. But still with these tools, you need a lot of typing when adding a reference. Also it seems that the abstract is lost when the reference is transfered to/from connotea (in RIS format). This is not the case for citeulike.

Then while browsing elsevier journals, I discovered 2collab. This later has been created by the same people behind sciencedirect and scopus meaning that you can import the full reference from these two services with one click. The full reference actually no; the abstract of the paper is truncated. Another nice thing, you can export all the references that you used in your papers as a ris file from scopus.

Finally, a friend of mine showed me zotero. This is really the ultimate tool that helps gather, organize and analyzes internet source. But it is a plugin for firefox and the created library is stored locally (perhaps this will change with the version 2.0). But the library can be exported as ris file (for instance) without any loss of information.

In summary, zotero is the best choice for me. I can always transfer the ris file exported from it to upload it to citeulike. I can also use import the ris file into JabRef since I use latex and bibtex to write papers.