As mentioned in Pete Clark's answer, you can absolutely arXiv your paper before submission, after submission, after acceptance (in the post-referee, author formated version), or after publication in any AMS journal.
Most publishers have equally liberal policies with respect to arXiving papers, which is summarized at SHERPA/RoMEO. However there are some pitfalls (which to the best of my knowledge have had no practical implication whatsoever) with some publishers.
Elsevier has an habit of mudding its policies, so it is safer to arXiv papers before submission, to retain the right to update it according the referee's requests.
Some publisher officially forbid the authors to publish the post-review version of a paper in repositories such as arXiv. This is the worst policy I know, as it implies forbidding to correct errors or inaccuracies in available preprints. Unfortunately, even some academic publishers such as Oxford University Press, engage in such nonsense. In practice, I did violate this clause of a copyright transfer agreement I signed, and nothing happened. Maybe they'll go after me know that I confessed.