I would profile both. But I would guess the 'LIKE' would be much faster, because it uses the binary search on the index (if the field is indexed). If you use the SUBSTR method, you will end up with a full table scan, as Oracle has to process row by row the function.