A probe recognition task requiring either identity, homophone, or synonym recognition was used to test an acoustic-to-semantic recoding hypothesis which had been suggested as an alternative explanation of a previous demonstration of semantic encoding in STM. With synonym recognition, conditions were compared where identical acoustic inputs were provided but in one condition a semantic relationship existed between the probe word and one of the list words while such a relationship did not exist in the other condition. The recoding hypothesis was not supported as the false-alarm rate in the latter condition remained at chance level.