To answer the question asked, the DateDiff will be more efficient than the convert. You might want to check out the "Common Date/Time Routines" link in my signature.

However, neither of these will be very efficient in a where clause... neither is SARGable, and neither will be able to utilize an index on the column. The best you can hope for with these is an index scan, though you may also get a table scan or clustered index scan. If this is a large table, just take a lunch break while it's running.

Elliott's suggestion to have a persisted, calculated column would work out here - just add an index to that column, and you'll be able to take care of both problems.