The approach I would use depends entirely on which codes are stored. Some need to be driven while observing the data stream (still don't clear the codes though unless it is a known intermittent), some are "pattern failures" that I just go right to the the test... Login to read more.