Difficult search tasks are known to involve attentional resources, but the spatio-temporal behavior of attention remains unknown. Are multiple search targets processed in sequence or in parallel? We developed a new methodology (based on (Dubois, Hamker and VanRullen, 2009, J. Vis. 9(5):3, 1-11)) to solve this notoriously difficult problem. Subjects (n=14) performed a difficult search task (detect a T among L letters) during which two additional probe letters were flashed at varying delays, at randomly determined locations previously occupied by two of the search items. Performance in reporting probes at each location was considered as a measure of attention deployment. By solving a second-degree equation, we determined the probability of probe report at the most and least attended probe locations on each trial. Because these values were not equal, we conclude that attention was focused on one stimulus or sub-group of stimuli at a time, and not divided equally between all search items. Furthermore, this focalization of attention (measured as the difference between most- and least-attended probe report performances) was modulated periodically over time at a frequency of ~7Hz. These results definitively rule out a strict parallel model of attention processing during this difficult search task. Instead, they suggest that attention focuses on a subset of items, and periodically samples the search array.