Participants read a series of passages containing an action that required the use of an instrument. In Experiment 1, a naming task failed to detect activation of a target instrument when that instrument was supported in the preceding text. In Experiment 2, reading times were slow on a target sentence that contradicted the inferential information, indicating that instrument information had been activated. The combination of the results from Experiments 1 and 2 are consistent with the view that activation involves a set of features of a concept, not a specific lexical item. The results from Experiment 3 and 4 demonstrated the role of contextual support in the activation and instantiation of instrumental inferences. (Contains 7 tables.)