The goal of the present research was to understand the relationship between object similarity and incidentally-inferred category structure. Participants completed one of two possible training tasks in which they learned to identify line-drawings of main battle tanks. In the “observation” group, participants observed a series of tank images and tank-model names. In the “response” group, participants saw the same tank images but had to guess the label on each trial (corrective feedback followed each response). All participants then completed an identification test, followed by a card-sorting task using the same tank images to determine how participants mentally organized the vehicles after repeated exposure. The response training condition resulted in significantly fewer identification errors than the observation training condition. The card sorts enabled us to predict a substantial proportion of the identification confusions, particularly after response training (50% of identification errors predicted). These results support the hypothesis that rote memorization of individual training items may lead to some (untrained) mental organization of items, and that identification errors may be partially predicted and explained as a result of this mental organization.