Visual information contributes fundamentally to the process of object categorization. The present study investigated whether the degree of activation of visual information in this process is dependent on the situational relevance of this information. We used the Proactive Interference (PI) paradigm. In two experiments, we manipulated the information by which objects could be retrieved from memory: by both semantic and shape information or by shape information only. The pattern of PI-release showed that if objects could be stored and retrieved both by semantic and shape information, then shape information was overruled by semantic information. If, however, semantic information could not be (satisfactorily) used to store and retrieve objects, then objects were stored in memory in terms of their shape.