Go to good conferences in your area of expertize to meet prospective collaborators - listen to their talks, and talk to them in person at coffee breaks, and exchange e-mail addresses. That's better - I think - than e-mail, as you get a feel for the real person behind the "science mask". If you are nor far away from each other, visit one another to test whether you can do something together, or otherwise make notes over - say - Google Docs - in order to exchange ideas, and schedule the work from that. If you do not "click", then try it again with someone else.
If you have more senior scientists to talk with at your university, then ask if they would know of junior scientists that have similar interests.
Don't do it! It will cramp your style and they can steal your data. It's happened to me. If you can publish on your own, it's more of an achievement and you can be more creative, even though it's harder.