I would do a profiler trace of the DB activity during the time of occurance. Any number of things could be happening from common locking to more obscure problems like iterative recompilation of your SP due to mangled statistics...
It's too hard to tell from the limited information provided.