We investigated whether icon arrays help to make deficits in participants' comprehension of health risk information salient, especially for participants with low graph literacy (ability to read and use information depicted in graphs). 240 participants (126 female; average age: 27 years) were recruited through Prolific Academic. We presented treatment risk scenarios with either numbers only or numbers with icon arrays, and participants chose between two treatments or an explicit I don't know‚ÄövÑvp option. We included letter strings as a cognitive load manipulation, and measures of graph literacy and numeracy. Contrary to hypotheses, only numeracy affected participants' comprehension, ˜ìv°\\(^2\\) = 16.85, \\(p\\) <.001, \\(f\\) = 0.10, and calibration, ˜ìv°\\(^2\\) = 20.06, \\(p\\) <.001, \\(f\\) = 0.11, with no significant interaction between graph condition (icon vs numbers) and graph literacy. Our findings call into question the ways in which icon arrays are beneficial for presenting risk information. Future research should further investigate how appropriate icon arrays are for presenting risk information, and explore the utility of an I don't know‚ÄövÑvp option in these scenarios.