Studies in visual category learning show that participants use different category generalization strategies. Some studies report a preference for a rule-based strategy, while others report a preference for a similarity-based strategy. We conducted category learning experiments in which we varied three variables - family resemblance of a category, saliency of the defining rule and presentation of transfer stimulus after a delay. Our results show that these factors influence the choice of category generalization strategy. Our study offers a possible explanation for the divergent results in the literature.