We had similar problems at a job I worked. Basically we had a third party software that the dev manager was impatient with the software producer. So he wrote his own applications against the db. Eventually the db hit a critical mass of usage, vs changes, vs triggers to contain the business logic.

Turns out the software in question always wrote data to tables in a consistent table order, found using profiler. Efforts were taken in development to make sure that all triggers and code followed the same order of inserts and updates to minimize the possibility of deadlocks. If you use the Profiler tool during these periods of deadlock you should be able to narrow down the processes in question that are causing the problem. Work out the data flow for these processes and you should be able to find your major culprit.