The experiment examines the effects of relatively long-term (10-40 minutes) priming on the recognition of spoken words presented in noise. The focus of the experiment was the extent to which the priming was lexical (i.e. whole word), morphological or physical. First, no priming was found for physically related words. Thus ‘deflecting’ spoken clearly in the pre-training task had no effect on the identification of ‘reflecting’ in the test phase. Second, there was very strong facilitation from a word with a regular inflectional relationship to the test word. Thus the word ‘reflected’ facilitated recognition of ‘reflecting’ almost as much as prior experience of ‘reflecting’ itself and, when responses were scored in terms of the root morpheme, counting ‘reflect’, ‘reflects’ and ‘reflected’ etc. as being correct, the morphemic effect was the same in the two cases. That this was a structural and not a semantic effect was shown by the complete lack of facilitation between irregularly related words such as ‘held/holding’, ‘man/men’, ‘lost/loses’. These data suggest that there is a morphological/structural level of analysis which is pre-semantic, at which these long-term facilitation effects take place. They also pose a challenge to the cohort theory of word recognition.