In this chapter, we show how visual attention can enhance the performance in object recognition. This proposition is inspired by the idea introduced in Neisser (Cognitive psychology. Appleton-CenturyCrofts, New York, 1967), which indicates that object recognition in human perception consists of two steps: an “attentional process selects the region of interest” and “complex object recognition processes are restricted to these regions”. Recently, in computer vision, much work has been done to combine the two domains of visual attention and object recognition, which we will call “attentive content-based image retrieval”. The test on our attentive CBIR approach in VOC 2005 demonstrated that we can maintain approximately the same recognition performance by selecting only 40 % of SIFT keypoints using classical saliency models. The proposed attentive CBIR framework can also be used to make a ranking between existing saliency models when used for CBIR. This ranking is different from the one using classical ground-truth like eye-tracking which means that choosing the best saliency models in predicting eye-tracking might be misleading when focusing on a CBIR application.