<p>High-level features of objects, such as meaning, bias attention even when task-irrelevant. We hypothesize that task-irrelevant semantic features bias attention via a grouping mechanism, organizing visual input by semantic relatedness in a manner similar to low-level grouping. Specifically, when presented with a task-irrelevant visual array of items, visual search is more efficient within a group of semantically related items. Participants were shown four stimuli of either color squares (low-level grouping), grayscale real-world objects (high-level grouping), or color real-world objects (low- and high-level grouping). On each trial two or three of the four items belonged to one category (e.g., clothing, blue squares). A target appeared randomly on one of the items, independent of relatedness (task-irrelevant). For all manipulations, search was equally efficient in groups of equal size. For unequal size groups, in color squares and grayscale objects, large groups yielded less efficient search – consistent with low-level perceptual grouping. In color objects, however, search was more efficient in a larger semantically related group – consistent with semantic bias. Results show that single features group displays, but complex stimuli bias attention beyond simple grouping.</p>