Try parsing the file 256 times — once using each ASCII character as the field delimiter — and collect statistics on the number of columns that result. The delimiter character that yields the most uniform number of fields per line is the likely one. (Yes, I've actually done this. Works pretty well sometimes.)