We introduce a two-person “beauty contest” game (aka guessing game) played on a two dimensional plane. Each player chooses a location and is rewarded by hitting a personal "target" based on his opponent's location. By observing subjects' eye movements on the graphs, we are able to observe the decision-making process. A level-k model of heterogeneous cognition can explain both the final choices and the lookup patterns during decision-making process. We fit a Markov-switching model to describe the lookup patterns, and classify subjects into various level-k types. Moreover, we employ a likelihood ratio test to select between competing level-k types, and show that when a similar likelihood ratio test cannot reject alternative level-k types using final choices alone, a further analysis of the lookups helps us distinguish between competing types. In particular, when the two classifications disagree, choice data alone could not reject nearly all of the lookup-based type alternatives, while most of the choice-based type alternatives could be rejected using lookup data.