This study examined the role of visual representation type, spatial ability, and reading comprehension in word problem solving in 128 sixth-grade students by using primarily an item-level approach rather than a test-level approach. We revealed that compared to students who did not make a visual representation, those who produced an accurate visual-schematic representation increased the chance of solving a word problem correctly almost six times. Inaccurate visual-schematic and pictorial representations, on the other hand, decreased students' chance of problem solving success. Noteworthy, reading comprehension was related to word problem solving at the test-level but not at the item-level. In interpreting the results, we advocate the use of item-level analyses since they are able to disclose such level-of-analysis discrepancies. © 2014.