This study tested the hypothesis that strategic and non-strategic learning result in knowledge representations that are distinguished by different modes of retrieval for information about constituent structure. Subjects were taught concepts under conditions designed to induce either a strategic or non-strategic mode of learning. Following training, subjects participated in an attribute priming task designed to assess the extent to which information about constituent structure was automatically or strategically activated when concept knowledge was retrieved. Activation was measured at three SOAs: 250 msec, 500 msec, and 2000 msec. For the strategically learned concept, information about constituent structure did not show evidence of activation until 2000 msec. For the non-strategically learned concept this information was activated by 250 msec but had declined to baseline by 2000 msec. These findings suggest that information about constituent structure is available for retrieval by strategic processes following strategic learning, but that this information is not available for retrieval by strategic processes following non-strategic learning. It is, however, available for automatic retrieval following non-strategic learning.