I'm looking at an old application that has an ID field for several of the tables stored in a keyseed table. When a new record is added, a process retrieves the value from the appropriate keyseed record, increments it, and updates the row. This was a fairly common technique at the time this app was written. Over the course of 10 years or so, there are probably only 3 or 4 examples where the process has failed to yield a unique value--but even that's too many. The fields are not all integers, some are actually stored as character data. They are used as candidate keys; the primary key is a SQL identity. Now that the data has been moved to SQL, what is the best way to insure this process yields no duplicates?