We compared the ability of two legend designs on a soil-landscape map to efficiently and effectively support map reading tasks with the goal of better understanding how the design choices affect user performance. Developing such knowledge is essential to design effective interfaces for digital earth systems. One of the two legends contained an alphabetical ordering of categories, while the other used a perceptual grouping based on the Munsell color space. We tested the two legends for 4 tasks with 20 experts (in geography-related domains). We analyzed traditional usability metrics and participants’ eye movements to identify the possible reasons behind their success and failure in the experimental tasks. Surprisingly, an overwhelming majority of the participants failed to arrive at the correct responses for two of the four tasks, irrespective of the legend design. Furthermore, participants’ prior knowledge of soils and map interpretation abilities led to interesting performance differences between the two legend types. We discuss how participant background might have played a role in performance and why some tasks were particularly hard to solve despite participants’ relatively high levels of experience in map reading. Based on our observations, we caution soil cartographers to be aware of the perceptual complexity of soil-landscape maps.