While the visual search task has been instrumental in the study of feature-based attention (i.e., how attentional mechanisms increase the salience of relevant features), no single methodology can tell the whole story. This dissertation investigates the contributions that a complementary methodology, the centroid task, can make to our understanding of feature-based attention. The first chapter directly compares the search and centroid tasks. It replicates the expected search results, finding that performance is always worse for conjunctive targets than it is for feature targets. However, it reports a different pattern of centroid results: conjunctive target centroid judgments can actually outperform constituent-feature centroid judgments. The second chapter examines the role of target-distractor similarity in centroid estimations. It finds that, given sufficiently salient feature contrasts, conjunctive target conditions are better than or equal to both constituent-feature conditions, suggesting that there is not necessarily any cost to conjunctive centroid judgments. The third chapter reviews an equisalience analysis of an 8-item centroid task and an analogous 2-item task. The different equisalience functions for the two tasks suggest that they access information differently. Together, these chapters provide compelling evidence that the centroid paradigm allows us to study aspects of feature-based attention that visual search cannot capture.