An experiment to determine the relative importance of the factors affecting recogniser performance with telephone speech is described. The design was controlled and balanced, and an ANOVA analysis was performed. Results are presented for single effects and interactions.