Summary: We present the outline of a closed-loop attention model (CLAM), to model attentional phenomena such as visual search. CLAM consists of four parts: retinotopic areas, feature maps, spatial maps, and working memory. These parts form a closed-loop diamond structure, with reciprocal connections. Visual search will result from the dynamic interactions in the diamond connection structure of CLAM. A target in working memory influences selection in the feature maps or the spatial maps, which then interact with the retinotopic areas. A target will be selected if there are not too many different distractors. Otherwise, a serial process will result, in which the interaction process in CLAM is limited to a restricted part of visual space.