Methodological issues in the assessment of commercial telephone-based speaker-independent speech recognisers are discussed, focussing on the practical decisions which must be made to arrive at a suitable standard. Results are presented for eight such state of the art recognisers tested on a fairly large UK telephone speech database.