The current study explored possible sources of demographic effects through analyses of errors from modified formats of the Benton Visual Retention Test (BVRT) completed by African American elders. Results indicate that: (1) reading level was a stronger predictor of BVRT performance than years of education; (2) on the single-item matching format of the task, individuals with lower reading levels disproportionately produced errors on items that differed in geometric, rather than spatial features; and (3) on a multiple-choice matching format, individuals with lower reading levels committed more errors on items where the target was located in the lower half of a 2x2 matrix.