Contrasts between constructed response items and stem-equivalent multiple-choice counterparts typically have involved averaging item characteristics, and this aggregation has masked differences in statistical properties at the item level. Moreover, even aggregated format differences have not been explained in terms of differential cognitive processing demands of the items. In this paper, item-level differences between figural response items and their multiple-choice counterparts in architecture are examined. The figural response item format is an assessment form that uses figural materials (such as graphs, illustrations, and diagrams) as item stimuli and the medium through which knowledge and skill are demonstrated. Item-level format differences in difficulty are examined, and then whether there are corresponding differences in the cognitive processing requirements of the items that can account for the psychometric differences is studied. Based on the evidence uncovered for these connections, it is proposed that differences in processing requirements and concomitant psychometric properties might be systematic and predictable. These analyses shed light on aspects of construct validity that are frequently neglected, and they touch the interface of the usually segregated psychometric and cognitive methodologies. Four tables and three figures illustrate the discussion. (Contains 20 references.) (Author/SLD)